CMSIS-DSP: Remove some gcc compilation warnings.

CMSIS/DSP/Examples/ARM/arm_bayes_example/arm_bayes_example_f32.c

@@ -92,6 +92,7 @@ int32_t main(void)
 
   /* Result of the classifier */
   float32_t result[NB_OF_CLASSES];
+  float32_t temp[NB_OF_CLASSES];
   float32_t maxProba;
   uint32_t index;
   
@@ -105,7 +106,7 @@ int32_t main(void)
   in[0] = 1.5f;
   in[1] = 1.0f;
 
-  index = arm_gaussian_naive_bayes_predict_f32(&S, in, result);
+  index = arm_gaussian_naive_bayes_predict_f32(&S, in, result,temp);
 
   maxProba = result[index];
 
@@ -116,7 +117,7 @@ int32_t main(void)
   in[0] = -1.5f;
   in[1] = 1.0f;
 
-  index = arm_gaussian_naive_bayes_predict_f32(&S, in, result);
+  index = arm_gaussian_naive_bayes_predict_f32(&S, in, result,temp);
 
   maxProba = result[index];
 
@@ -127,7 +128,7 @@ int32_t main(void)
   in[0] = 0.0f;
   in[1] = -3.0f;
 
-  index = arm_gaussian_naive_bayes_predict_f32(&S, in, result);
+  index = arm_gaussian_naive_bayes_predict_f32(&S, in, result,temp);
 
   maxProba = result[index];
 

CMSIS/DSP/Include/dsp/bayes_functions.h

@@ -67,9 +67,10 @@ typedef struct
 /**
  * @brief Naive Gaussian Bayesian Estimator
  *
- * @param[in]  S         points to a naive bayes instance structure
- * @param[in]  in        points to the elements of the input vector.
- * @param[in]  pBuffer   points to a buffer of length numberOfClasses
+ * @param[in]  S                        points to a naive bayes instance structure
+ * @param[in]  in                       points to the elements of the input vector.
+ * @param[out] *pOutputProbabilities    points to a buffer of length numberOfClasses containing estimated probabilities
+ * @param[out] *pBufferB                points to a temporary buffer of length numberOfClasses
  * @return The predicted class
  *
  */
@@ -77,7 +78,8 @@ typedef struct
 
 uint32_t arm_gaussian_naive_bayes_predict_f32(const arm_gaussian_naive_bayes_instance_f32 *S, 
    const float32_t * in, 
-   float32_t *pBuffer);
+   float32_t *pOutputProbabilities,
+   float32_t *pBufferB);
 
 
 #ifdef   __cplusplus

CMSIS/DSP/Include/dsp/bayes_functions_f16.h

@@ -58,9 +58,10 @@ typedef struct
 /**
  * @brief Naive Gaussian Bayesian Estimator
  *
- * @param[in]  S         points to a naive bayes instance structure
- * @param[in]  in        points to the elements of the input vector.
- * @param[in]  pBuffer   points to a buffer of length numberOfClasses
+ * @param[in]  S                        points to a naive bayes instance structure
+ * @param[in]  in                       points to the elements of the input vector.
+ * @param[out] *pOutputProbabilities    points to a buffer of length numberOfClasses containing estimated probabilities
+ * @param[out] *pBufferB                points to a temporary buffer of length numberOfClasses
  * @return The predicted class
  *
  */
@@ -68,7 +69,8 @@ typedef struct
 
 uint32_t arm_gaussian_naive_bayes_predict_f16(const arm_gaussian_naive_bayes_instance_f16 *S, 
    const float16_t * in, 
-   float16_t *pBuffer);
+   float16_t *pOutputProbabilities,
+   float16_t *pBufferB);
 
 #endif /*defined(ARM_FLOAT16_SUPPORTED)*/
 #ifdef   __cplusplus

CMSIS/DSP/Source/BayesFunctions/arm_gaussian_naive_bayes_predict_f16.c

@@ -43,13 +43,12 @@
 /**
  * @brief Naive Gaussian Bayesian Estimator
  *
- * @param[in]  *S         points to a naive bayes instance structure
- * @param[in]  *in        points to the elements of the input vector.
- * @param[in]  *pBuffer   points to a buffer of length numberOfClasses
+ * @param[in]  *S                       points to a naive bayes instance structure
+ * @param[in]  *in                      points to the elements of the input vector.
+ * @param[out] *pOutputProbabilities    points to a buffer of length numberOfClasses containing estimated probabilities
+ * @param[out] *pBufferB                points to a temporary buffer of length numberOfClasses
  * @return The predicted class
  *
- * @par If the number of classes is big, MVE version will consume lot of
- * stack since the log prior are computed on the stack.
  *
  */
 
@@ -60,19 +59,21 @@
 
 uint32_t arm_gaussian_naive_bayes_predict_f16(const arm_gaussian_naive_bayes_instance_f16 *S, 
    const float16_t * in, 
-   float16_t *pBuffer)
+   float16_t *pOutputProbabilities,
+   float16_t *pBufferB
+   )
 {
     uint32_t         nbClass;
     const float16_t *pTheta = S->theta;
     const float16_t *pSigma = S->sigma;
-    float16_t      *buffer = pBuffer;
+    float16_t      *buffer = pOutputProbabilities;
     const float16_t *pIn = in;
     float16_t       result;
     f16x8_t         vsigma;
     _Float16       tmp;
     f16x8_t         vacc1, vacc2;
     uint32_t        index;
-    float16_t       logclassPriors[S->numberOfClasses];
+    float16_t       *logclassPriors=pBufferB;
     float16_t      *pLogPrior = logclassPriors;
 
     arm_vlog_f16((float16_t *) S->classPriors, logclassPriors, S->numberOfClasses);
@@ -135,38 +136,31 @@ uint32_t arm_gaussian_naive_bayes_predict_f16(const arm_gaussian_naive_bayes_ins
         buffer++;
     }
 
-    arm_max_f16(pBuffer, S->numberOfClasses, &result, &index);
+    arm_max_f16(pOutputProbabilities, S->numberOfClasses, &result, &index);
 
     return (index);
 }
 
 #else
 
-/**
- * @brief Naive Gaussian Bayesian Estimator
- *
- * @param[in]  *S         points to a naive bayes instance structure
- * @param[in]  *in        points to the elements of the input vector.
- * @param[in]  *pBuffer   points to a buffer of length numberOfClasses
- * @return The predicted class
- *
- */
 uint32_t arm_gaussian_naive_bayes_predict_f16(const arm_gaussian_naive_bayes_instance_f16 *S, 
    const float16_t * in, 
-   float16_t *pBuffer)
+   float16_t *pOutputProbabilities,
+   float16_t *pBufferB)
 {
     uint32_t nbClass;
     uint32_t nbDim;
     const float16_t *pPrior = S->classPriors;
     const float16_t *pTheta = S->theta;
     const float16_t *pSigma = S->sigma;
-    float16_t *buffer = pBuffer;
+    float16_t *buffer = pOutputProbabilities;
     const float16_t *pIn=in;
     float16_t result;
     _Float16 sigma;
     _Float16 tmp;
     _Float16 acc1,acc2;
     uint32_t index;
+    (void)pBufferB;
 
     pTheta=S->theta;
     pSigma=S->sigma;
@@ -199,7 +193,7 @@ uint32_t arm_gaussian_naive_bayes_predict_f16(const arm_gaussian_naive_bayes_ins
         buffer++;
     }
 
-    arm_max_f16(pBuffer,S->numberOfClasses,&result,&index);
+    arm_max_f16(pOutputProbabilities,S->numberOfClasses,&result,&index);
 
     return(index);
 }

CMSIS/DSP/Source/BayesFunctions/arm_gaussian_naive_bayes_predict_f32.c

@@ -41,13 +41,12 @@
 /**
  * @brief Naive Gaussian Bayesian Estimator
  *
- * @param[in]  *S         points to a naive bayes instance structure
- * @param[in]  *in        points to the elements of the input vector.
- * @param[in]  *pBuffer   points to a buffer of length numberOfClasses
+ * @param[in]   *S                      points to a naive bayes instance structure
+ * @param[in]   *in                     points to the elements of the input vector.
+ * @param[out]  *pOutputProbabilities   points to a buffer of length numberOfClasses containing estimated probabilities
+ * @param[out]  *pBufferB               points to a temporary buffer of length numberOfClasses
  * @return The predicted class
  *
- * @par If the number of classes is big, MVE version will consume lot of
- * stack since the log prior are computed on the stack.
  *
  */
 
@@ -58,19 +57,21 @@
 
 uint32_t arm_gaussian_naive_bayes_predict_f32(const arm_gaussian_naive_bayes_instance_f32 *S, 
    const float32_t * in, 
-   float32_t *pBuffer)
+   float32_t *pOutputProbabilities,
+   float32_t *pBufferB
+   )
 {
     uint32_t         nbClass;
     const float32_t *pTheta = S->theta;
     const float32_t *pSigma = S->sigma;
-    float32_t      *buffer = pBuffer;
+    float32_t      *buffer = pOutputProbabilities;
     const float32_t *pIn = in;
     float32_t       result;
     f32x4_t         vsigma;
     float32_t       tmp;
     f32x4_t         vacc1, vacc2;
     uint32_t        index;
-    float32_t       logclassPriors[S->numberOfClasses];
+    float32_t       *logclassPriors=pBufferB;
     float32_t      *pLogPrior = logclassPriors;
 
     arm_vlog_f32((float32_t *) S->classPriors, logclassPriors, S->numberOfClasses);
@@ -133,7 +134,7 @@ uint32_t arm_gaussian_naive_bayes_predict_f32(const arm_gaussian_naive_bayes_ins
         buffer++;
     }
 
-    arm_max_f32(pBuffer, S->numberOfClasses, &result, &index);
+    arm_max_f32(pOutputProbabilities, S->numberOfClasses, &result, &index);
 
     return (index);
 }
@@ -148,7 +149,8 @@ uint32_t arm_gaussian_naive_bayes_predict_f32(const arm_gaussian_naive_bayes_ins
 
 uint32_t arm_gaussian_naive_bayes_predict_f32(const arm_gaussian_naive_bayes_instance_f32 *S, 
    const float32_t * in, 
-   float32_t *pBuffer)
+   float32_t *pOutputProbabilities,
+   float32_t *pBufferB)
 {
     
     const float32_t *pPrior = S->classPriors;
@@ -159,7 +161,7 @@ uint32_t arm_gaussian_naive_bayes_predict_f32(const arm_gaussian_naive_bayes_ins
     const float32_t *pTheta1 = S->theta + S->vectorDimension;
     const float32_t *pSigma1 = S->sigma + S->vectorDimension;
 
-    float32_t *buffer = pBuffer;
+    float32_t *buffer = pOutputProbabilities;
     const float32_t *pIn=in;
 
     float32_t result;
@@ -174,6 +176,7 @@ uint32_t arm_gaussian_naive_bayes_predict_f32(const arm_gaussian_naive_bayes_ins
     float32x2_t tmpV2;
     float32x4_t thetaV,thetaV1;
     float32x4_t inV;
+    (void)pBufferB;
 
     epsilonV = vdupq_n_f32(S->epsilon);
 
@@ -322,32 +325,24 @@ uint32_t arm_gaussian_naive_bayes_predict_f32(const arm_gaussian_naive_bayes_ins
         classBlkCnt--;
     }
 
-    arm_max_f32(pBuffer,S->numberOfClasses,&result,&index);
+    arm_max_f32(pOutputProbabilities,S->numberOfClasses,&result,&index);
 
     return(index);
 }
 
 #else
 
-/**
- * @brief Naive Gaussian Bayesian Estimator
- *
- * @param[in]  *S         points to a naive bayes instance structure
- * @param[in]  *in        points to the elements of the input vector.
- * @param[in]  *pBuffer   points to a buffer of length numberOfClasses
- * @return The predicted class
- *
- */
 uint32_t arm_gaussian_naive_bayes_predict_f32(const arm_gaussian_naive_bayes_instance_f32 *S, 
    const float32_t * in, 
-   float32_t *pBuffer)
+   float32_t *pOutputProbabilities,
+   float32_t *pBufferB)
 {
     uint32_t nbClass;
     uint32_t nbDim;
     const float32_t *pPrior = S->classPriors;
     const float32_t *pTheta = S->theta;
     const float32_t *pSigma = S->sigma;
-    float32_t *buffer = pBuffer;
+    float32_t *buffer = pOutputProbabilities;
     const float32_t *pIn=in;
     float32_t result;
     float32_t sigma;
@@ -355,6 +350,8 @@ uint32_t arm_gaussian_naive_bayes_predict_f32(const arm_gaussian_naive_bayes_ins
     float32_t acc1,acc2;
     uint32_t index;
 
+    (void)pBufferB;
+
     pTheta=S->theta;
     pSigma=S->sigma;
 
@@ -386,7 +383,7 @@ uint32_t arm_gaussian_naive_bayes_predict_f32(const arm_gaussian_naive_bayes_ins
         buffer++;
     }
 
-    arm_max_f32(pBuffer,S->numberOfClasses,&result,&index);
+    arm_max_f32(pOutputProbabilities,S->numberOfClasses,&result,&index);
 
     return(index);
 }

CMSIS/DSP/Source/ComplexMathFunctions/arm_cmplx_mult_real_f16.c

@@ -79,7 +79,7 @@ void arm_cmplx_mult_real_f16(
         float16_t * pCmplxDst,
         uint32_t numSamples)
 {
-    const static uint16_t stride_cmplx_x_real_16[8] = {
+    static const uint16_t stride_cmplx_x_real_16[8] = {
         0, 0, 1, 1, 2, 2, 3, 3
         };
     uint32_t blockSizeC = numSamples * CMPLX_DIM;   /* loop counters */

CMSIS/DSP/Source/ComplexMathFunctions/arm_cmplx_mult_real_f32.c

@@ -77,7 +77,7 @@ void arm_cmplx_mult_real_f32(
         float32_t * pCmplxDst,
         uint32_t numSamples)
 {
-    const static uint32_t stride_cmplx_x_real_32[4] = { 0, 0, 1, 1 };
+    static const uint32_t stride_cmplx_x_real_32[4] = { 0, 0, 1, 1 };
 
     uint32_t blockSizeC = numSamples * CMPLX_DIM;   /* loop counters */
     uint32_t blkCnt;

CMSIS/DSP/Source/ComplexMathFunctions/arm_cmplx_mult_real_q15.c

@@ -57,7 +57,7 @@ void arm_cmplx_mult_real_q15(
         q15_t * pCmplxDst,
         uint32_t numSamples)
 {
-  const static uint16_t stride_cmplx_x_real_16[8] = {
+  static const uint16_t stride_cmplx_x_real_16[8] = {
       0, 0, 1, 1, 2, 2, 3, 3
       };
   q15x8_t rVec;

CMSIS/DSP/Source/ComplexMathFunctions/arm_cmplx_mult_real_q31.c

@@ -58,7 +58,7 @@ void arm_cmplx_mult_real_q31(
         uint32_t numSamples)
 {
 
-    const static uint32_t stride_cmplx_x_real_32[4] = {
+    static const uint32_t stride_cmplx_x_real_32[4] = {
         0, 0, 1, 1
     };
     q31x4_t rVec;

CMSIS/DSP/Source/FilteringFunctions/arm_conv_partial_f32.c

@@ -142,7 +142,7 @@ arm_status arm_conv_partial_f32(
     blockSize3 = ((int32_t)check > (int32_t)srcALen) ? (int32_t)check - (int32_t)srcALen : 0;
     blockSize3 = ((int32_t)firstIndex > (int32_t)srcALen - 1) ? blockSize3 - (int32_t)firstIndex + (int32_t)srcALen : blockSize3;
     blockSize1 = ((int32_t) srcBLen - 1) - (int32_t) firstIndex;
-    blockSize1 = (blockSize1 > 0) ? ((check > (srcBLen - 1U)) ? blockSize1 : numPoints) : 0;
+    blockSize1 = (blockSize1 > 0) ? ((check > (srcBLen - 1U)) ? blockSize1 : (int32_t)numPoints) : 0;
     blockSize2 = ((int32_t) check - blockSize3) - (blockSize1 + (int32_t) firstIndex);
     blockSize2 = (blockSize2 > 0) ? blockSize2 : 0;
 

CMSIS/DSP/Source/FilteringFunctions/arm_conv_partial_fast_q31.c

@@ -118,7 +118,7 @@ arm_status arm_conv_partial_fast_q31(
     blockSize3 = ((int32_t)check > (int32_t)srcALen) ? (int32_t)check - (int32_t)srcALen : 0;
     blockSize3 = ((int32_t)firstIndex > (int32_t)srcALen - 1) ? blockSize3 - (int32_t)firstIndex + (int32_t)srcALen : blockSize3;
     blockSize1 = ((int32_t) srcBLen - 1) - (int32_t) firstIndex;
-    blockSize1 = (blockSize1 > 0) ? ((check > (srcBLen - 1U)) ? blockSize1 :  numPoints) : 0;
+    blockSize1 = (blockSize1 > 0) ? ((check > (srcBLen - 1U)) ? blockSize1 :  (int32_t)numPoints) : 0;
     blockSize2 = (int32_t) check - ((blockSize3 + blockSize1) + (int32_t) firstIndex);
     blockSize2 = (blockSize2 > 0) ? blockSize2 : 0;
 

CMSIS/DSP/Source/FilteringFunctions/arm_conv_partial_q15.c

@@ -119,7 +119,7 @@ arm_status arm_conv_partial_q15(
     blockSize3 = ((int32_t)check > (int32_t)srcALen) ? (int32_t)check - (int32_t)srcALen : 0;
     blockSize3 = ((int32_t)firstIndex > (int32_t)srcALen - 1) ? blockSize3 - (int32_t)firstIndex + (int32_t)srcALen : blockSize3;
     blockSize1 = ((int32_t) srcBLen - 1) - (int32_t) firstIndex;
-    blockSize1 = (blockSize1 > 0) ? ((check > (srcBLen - 1U)) ? blockSize1 :  numPoints) : 0;
+    blockSize1 = (blockSize1 > 0) ? ((check > (srcBLen - 1U)) ? blockSize1 :  (int32_t)numPoints) : 0;
     blockSize2 = (int32_t) check - ((blockSize3 + blockSize1) + (int32_t) firstIndex);
     blockSize2 = (blockSize2 > 0) ? blockSize2 : 0;
 

CMSIS/DSP/Source/FilteringFunctions/arm_conv_partial_q31.c

@@ -121,7 +121,7 @@ arm_status arm_conv_partial_q31(
     blockSize3 = ((int32_t)check > (int32_t)srcALen) ? (int32_t)check - (int32_t)srcALen : 0;
     blockSize3 = ((int32_t)firstIndex > (int32_t)srcALen - 1) ? blockSize3 - (int32_t)firstIndex + (int32_t)srcALen : blockSize3;
     blockSize1 = ((int32_t) srcBLen - 1) - (int32_t) firstIndex;
-    blockSize1 = (blockSize1 > 0) ? ((check > (srcBLen - 1U)) ? blockSize1 :  numPoints) : 0;
+    blockSize1 = (blockSize1 > 0) ? ((check > (srcBLen - 1U)) ? blockSize1 :  (int32_t)numPoints) : 0;
     blockSize2 = (int32_t) check - ((blockSize3 + blockSize1) + (int32_t) firstIndex);
     blockSize2 = (blockSize2 > 0) ? blockSize2 : 0;
 

CMSIS/DSP/Source/FilteringFunctions/arm_conv_partial_q7.c

@@ -123,7 +123,7 @@ arm_status arm_conv_partial_q7(
     blockSize3 = ((int32_t)check > (int32_t)srcALen) ? (int32_t)check - (int32_t)srcALen : 0;
     blockSize3 = ((int32_t)firstIndex > (int32_t)srcALen - 1) ? blockSize3 - (int32_t)firstIndex + (int32_t)srcALen : blockSize3;
     blockSize1 = ((int32_t) srcBLen - 1) - (int32_t) firstIndex;
-    blockSize1 = (blockSize1 > 0) ? ((check > (srcBLen - 1U)) ? blockSize1 : numPoints) : 0;
+    blockSize1 = (blockSize1 > 0) ? ((check > (srcBLen - 1U)) ? blockSize1 : (int32_t)numPoints) : 0;
     blockSize2 = (int32_t) check - ((blockSize3 + blockSize1) + (int32_t) firstIndex);
     blockSize2 = (blockSize2 > 0) ? blockSize2 : 0;
 

CMSIS/DSP/Source/FilteringFunctions/arm_fir_f16.c

@@ -59,6 +59,7 @@
             vecAcc0 = vfmaq(vecAcc0, vecIn0, c[i]);                        \
         }
 
+#define NB_TAPS 4
 __STATIC_INLINE void arm_fir_f16_1_4_mve(const arm_fir_instance_f16 * S, 
     const float16_t * __restrict pSrc, 
     float16_t * __restrict pDst, uint32_t blockSize)
@@ -74,7 +75,6 @@ __STATIC_INLINE void arm_fir_f16_1_4_mve(const arm_fir_instance_f16 * S,
     int32_t         blkCnt;
     float16x8_t         vecIn0;
     float16x8_t         vecAcc0;
-    const int       NB_TAPS=4;
     float16_t       c[NB_TAPS];
 
 
@@ -147,8 +147,9 @@ __STATIC_INLINE void arm_fir_f16_1_4_mve(const arm_fir_instance_f16 * S,
     }
 
 }
+#undef NB_TAPS
 
-
+#define NB_TAPS 8
 __STATIC_INLINE void arm_fir_f16_5_8_mve(const arm_fir_instance_f16 * S, 
     const float16_t * __restrict pSrc, 
     float16_t * __restrict pDst, uint32_t blockSize)
@@ -164,7 +165,6 @@ __STATIC_INLINE void arm_fir_f16_5_8_mve(const arm_fir_instance_f16 * S,
     int32_t         blkCnt;
     float16x8_t         vecIn0;
     float16x8_t         vecAcc0;
-    const int       NB_TAPS=8;
     float16_t       c[NB_TAPS];
 
 
@@ -237,7 +237,7 @@ __STATIC_INLINE void arm_fir_f16_5_8_mve(const arm_fir_instance_f16 * S,
     }
 
 }
-
+#undef NB_TAPS
 
 void arm_fir_f16(const arm_fir_instance_f16 * S, 
   const float16_t * pSrc, 

CMSIS/DSP/Source/FilteringFunctions/arm_fir_f32.c

@@ -160,6 +160,7 @@
         }
 
 
+#define NB_TAPS 4
 __STATIC_INLINE void arm_fir_f32_1_4_mve(const arm_fir_instance_f32 * S, 
   const float32_t * __restrict pSrc, 
   float32_t * __restrict pDst, uint32_t blockSize)
@@ -176,7 +177,6 @@ __STATIC_INLINE void arm_fir_f32_1_4_mve(const arm_fir_instance_f32 * S,
     int32_t         blkCnt;
     float32x4_t         vecIn0;
     float32x4_t         vecAcc0;
-    const int       NB_TAPS=4;
     float32_t       c[NB_TAPS];
     const float32_t *pCoeffsCur = pCoeffs;
 
@@ -243,8 +243,7 @@ __STATIC_INLINE void arm_fir_f32_1_4_mve(const arm_fir_instance_f32 * S,
     }
     while (blkCnt > 0);
 }
-
-
+#undef NB_TAPS
 
 __STATIC_INLINE void arm_fir_f32_5_8_mve(const arm_fir_instance_f32 * S, 
   const float32_t * __restrict pSrc, 

CMSIS/DSP/Source/FilteringFunctions/arm_fir_q15.c

@@ -86,14 +86,13 @@
     uint32_t        numTaps = S->numTaps;   /* Number of filter coefficients in the filter */\
     int32_t         blkCnt;                                                                  \
     q15x8_t         vecIn0;                                                                  \
-    const int32_t   nbVecTaps = (NBTAPS / 8);                                                \
                                                                                              \
     /*                                                                                       \
      * load coefs                                                                            \
      */                                                                                      \
-    q15x8_t         vecCoeffs[nbVecTaps];                                                    \
+    q15x8_t         vecCoeffs[NBVECTAPS];                                                    \
                                                                                              \
-    for (int i = 0; i < nbVecTaps; i++)                                                      \
+    for (int i = 0; i < NBVECTAPS; i++)                                                      \
         vecCoeffs[i] = vldrhq_s16(pCoeffs + 8 * i);                                          \
                                                                                              \
     /*                                                                                       \
@@ -114,7 +113,7 @@
         pStateCur += 4;                                                                      \
         pTempSrc += 4;                                                                       \
                                                                                              \
-        FIR_Q15_CORE(pOutput, 4, nbVecTaps, pSamples, vecCoeffs);                            \
+        FIR_Q15_CORE(pOutput, 4, NBVECTAPS, pSamples, vecCoeffs);                            \
         pSamples += 4;                                                                       \
                                                                                              \
         blkCnt--;                                                                            \
@@ -126,7 +125,7 @@
     for (int i = 0; i < residual; i++)                                                       \
         *pStateCur++ = *pTempSrc++;                                                          \
                                                                                              \
-    FIR_Q15_CORE(pOutput, residual, nbVecTaps, pSamples, vecCoeffs);                         \
+    FIR_Q15_CORE(pOutput, residual, NBVECTAPS, pSamples, vecCoeffs);                         \
                                                                                              \
     /*                                                                                       \
      * Copy the samples back into the history buffer start                                   \
@@ -156,7 +155,9 @@ static void arm_fir_q15_25_32_mve(const arm_fir_instance_q15 * S,
   q15_t * __restrict pDst, uint32_t blockSize)
 {
     #define NBTAPS 32
+    #define NBVECTAPS (NBTAPS / 8)
     FIR_Q15_MAIN_CORE();
+    #undef NBVECTAPS
     #undef NBTAPS
 }
 
@@ -165,7 +166,9 @@ static void arm_fir_q15_17_24_mve(const arm_fir_instance_q15 * S,
   q15_t * __restrict pDst, uint32_t blockSize)
 {
     #define NBTAPS 24
+    #define NBVECTAPS (NBTAPS / 8)
     FIR_Q15_MAIN_CORE();
+    #undef NBVECTAPS
     #undef NBTAPS
 }
 
@@ -175,7 +178,9 @@ static void arm_fir_q15_9_16_mve(const arm_fir_instance_q15 * S,
   q15_t * __restrict pDst, uint32_t blockSize)
 {
     #define NBTAPS 16
+    #define NBVECTAPS (NBTAPS / 8)
     FIR_Q15_MAIN_CORE();
+    #undef NBVECTAPS
     #undef NBTAPS
 }
 
@@ -184,7 +189,9 @@ static void arm_fir_q15_1_8_mve(const arm_fir_instance_q15 * S,
   q15_t * __restrict pDst, uint32_t blockSize)
 {
     #define NBTAPS 8
+    #define NBVECTAPS (NBTAPS / 8)
     FIR_Q15_MAIN_CORE();
+    #undef NBVECTAPS
     #undef NBTAPS
 }
 

CMSIS/DSP/Source/FilteringFunctions/arm_fir_q31.c

@@ -117,14 +117,13 @@
     q31_t       *pTempDest;             /* Temporary pointer to the destination buffer */\
     uint32_t     numTaps = S->numTaps;  /* Number of filter coefficients in the filter */\
     int32_t      blkCnt;                                                                 \
-    const int32_t   nbVecTaps = (NBTAPS / 4);                                            \
                                                                                          \
     /*                                                                                   \
      * load coefs                                                                        \
      */                                                                                  \
-    q31x4_t         vecCoeffs[nbVecTaps];                                                \
+    q31x4_t         vecCoeffs[NBVECTAPS];                                                \
                                                                                          \
-    for (int i = 0; i < nbVecTaps; i++)                                                  \
+    for (int i = 0; i < NBVECTAPS; i++)                                                  \
         vecCoeffs[i] = vld1q(pCoeffs + 4 * i);                                           \
                                                                                          \
     /*                                                                                   \
@@ -145,7 +144,7 @@
         pStateCur += 4;                                                                  \
         pTempSrc += 4;                                                                   \
                                                                                          \
-        FIR_Q31_CORE(4, nbVecTaps, pSamples, vecCoeffs);                                 \
+        FIR_Q31_CORE(4, NBVECTAPS, pSamples, vecCoeffs);                                 \
                                                                                          \
         pSamples += 4;                                                                   \
         /*                                                                               \
@@ -162,7 +161,7 @@
               for (int i = 0; i < residual; i++)                                         \
                   *pStateCur++ = *pTempSrc++;                                            \
                                                                                          \
-              FIR_Q31_CORE(3, nbVecTaps, pSamples, vecCoeffs);                           \
+              FIR_Q31_CORE(3, NBVECTAPS, pSamples, vecCoeffs);                           \
           }                                                                              \
           break;                                                                         \
                                                                                          \
@@ -171,7 +170,7 @@
               for (int i = 0; i < residual; i++)                                         \
                   *pStateCur++ = *pTempSrc++;                                            \
                                                                                          \
-               FIR_Q31_CORE(2, nbVecTaps, pSamples, vecCoeffs);                          \
+               FIR_Q31_CORE(2, NBVECTAPS, pSamples, vecCoeffs);                          \
           }                                                                              \
           break;                                                                         \
                                                                                          \
@@ -180,7 +179,7 @@
               for (int i = 0; i < residual; i++)                                         \
                   *pStateCur++ = *pTempSrc++;                                            \
                                                                                          \
-              FIR_Q31_CORE(1, nbVecTaps, pSamples, vecCoeffs);                           \
+              FIR_Q31_CORE(1, NBVECTAPS, pSamples, vecCoeffs);                           \
           }                                                                              \
           break;                                                                         \
     }                                                                                    \
@@ -382,7 +381,9 @@ static void arm_fir_q31_5_8_mve(const arm_fir_instance_q31 * S,
     q31_t * __restrict pDst, uint32_t blockSize)
 {
     #define NBTAPS 8
+    #define NBVECTAPS (NBTAPS / 4)
     FIR_Q31_MAIN_CORE();
+    #undef NBVECTAPS
     #undef NBTAPS
 }
 
@@ -392,7 +393,9 @@ static void arm_fir_q31_9_12_mve(const arm_fir_instance_q31 * S,
     q31_t * __restrict pDst, uint32_t blockSize)
 {
     #define NBTAPS 12
+    #define NBVECTAPS (NBTAPS / 4)
     FIR_Q31_MAIN_CORE();
+    #undef NBVECTAPS
     #undef NBTAPS
 }
 
@@ -402,7 +405,9 @@ static void arm_fir_q31_13_16_mve(const arm_fir_instance_q31 * S,
     q31_t * __restrict pDst, uint32_t blockSize)
 {
     #define NBTAPS 16
+    #define NBVECTAPS (NBTAPS / 4)
     FIR_Q31_MAIN_CORE();
+    #undef NBVECTAPS
     #undef NBTAPS
 }
 
@@ -412,7 +417,9 @@ static void arm_fir_q31_17_20_mve(const arm_fir_instance_q31 * S,
     q31_t * __restrict pDst, uint32_t blockSize)
 {
     #define NBTAPS 20
+    #define NBVECTAPS (NBTAPS / 4)
     FIR_Q31_MAIN_CORE();
+    #undef NBVECTAPS
     #undef NBTAPS
 }
 
@@ -422,7 +429,9 @@ static void arm_fir_q31_21_24_mve(const arm_fir_instance_q31 * S,
     q31_t * __restrict pDst, uint32_t blockSize)
 {
     #define NBTAPS 24
+    #define NBVECTAPS (NBTAPS / 4)
     FIR_Q31_MAIN_CORE();
+    #undef NBVECTAPS
     #undef NBTAPS
 }
 
@@ -432,7 +441,9 @@ static void arm_fir_q31_25_28_mve(const arm_fir_instance_q31 * S,
     q31_t * __restrict pDst, uint32_t blockSize)
 {
     #define NBTAPS 28
+    #define NBVECTAPS (NBTAPS / 4)
     FIR_Q31_MAIN_CORE();
+    #undef NBVECTAPS
     #undef NBTAPS
 }
 

CMSIS/DSP/Source/FilteringFunctions/arm_fir_q7.c

@@ -81,14 +81,13 @@
     uint32_t       numTaps = S->numTaps;   /* Number of filter coefficients in the filter */\
     int32_t        blkCnt;                                                                  \
     q7x16_t        vecIn0;                                                                  \
-    const int32_t  nbVecTaps = (NBTAPS / 16);                                     \
                                                                                             \
     /*                                                                                      \
      * load coefs                                                                           \
      */                                                                                     \
-    q7x16_t         vecCoeffs[nbVecTaps];                                                   \
+    q7x16_t         vecCoeffs[NBVECTAPS];                                                   \
                                                                                             \
-    for (int i = 0; i < nbVecTaps; i++)                                                     \
+    for (int i = 0; i < NBVECTAPS; i++)                                                     \
         vecCoeffs[i] = vldrbq_s8(pCoeffs + 16 * i);                               \
                                                                                             \
     /*                                                                                      \
@@ -109,7 +108,7 @@
         pStateCur += 4;                                                                     \
         pTempSrc += 4;                                                                      \
                                                                                             \
-        FIR_Q7_CORE(pOutput, 4, nbVecTaps, pSamples, vecCoeffs);                            \
+        FIR_Q7_CORE(pOutput, 4, NBVECTAPS, pSamples, vecCoeffs);                            \
         pSamples += 4;                                                                      \
                                                                                             \
         blkCnt--;                                                                           \
@@ -121,7 +120,7 @@
     for (int i = 0; i < residual; i++)                                                      \
         *pStateCur++ = *pTempSrc++;                                                         \
                                                                                             \
-    FIR_Q7_CORE(pOutput, residual, nbVecTaps, pSamples, vecCoeffs);                         \
+    FIR_Q7_CORE(pOutput, residual, NBVECTAPS, pSamples, vecCoeffs);                         \
                                                                                             \
                                                                                             \
     /*                                                                                      \
@@ -147,7 +146,9 @@ static void arm_fir_q7_49_64_mve(const arm_fir_instance_q7 * S,
   q7_t * __restrict pDst, uint32_t blockSize)
 {
     #define NBTAPS 64
+    #define NBVECTAPS (NBTAPS / 16)
     FIR_Q7_MAIN_CORE();
+    #undef NBVECTAPS
     #undef NBTAPS
 }
 
@@ -157,7 +158,9 @@ void arm_fir_q7_33_48_mve(const arm_fir_instance_q7 * S,
   q7_t * __restrict pDst, uint32_t blockSize)
 {
     #define NBTAPS 48
+    #define NBVECTAPS (NBTAPS / 16)
     FIR_Q7_MAIN_CORE();
+    #undef NBVECTAPS
     #undef NBTAPS
 }
 
@@ -166,7 +169,9 @@ static void arm_fir_q7_17_32_mve(const arm_fir_instance_q7 * S,
   q7_t * __restrict pDst, uint32_t blockSize)
 {
     #define NBTAPS 32
+    #define NBVECTAPS (NBTAPS / 16)
     FIR_Q7_MAIN_CORE();
+    #undef NBVECTAPS
     #undef NBTAPS
 }
 
@@ -176,7 +181,9 @@ void arm_fir_q7_1_16_mve(const arm_fir_instance_q7 * S,
   q7_t * __restrict pDst, uint32_t blockSize)
 {
     #define NBTAPS 16
+    #define NBVECTAPS (NBTAPS / 16)
     FIR_Q7_MAIN_CORE();
+    #undef NBVECTAPS
     #undef NBTAPS
 }
 

CMSIS/DSP/Testing/Include/Benchmarks/BayesF16.h

@@ -17,6 +17,7 @@ class BayesF16:public Client::Suite
             Client::Pattern<int16_t> dims;
 
             Client::LocalPattern<float16_t> outputProbas;
+            Client::LocalPattern<float16_t> temp;
             Client::LocalPattern<int16_t> outputPredicts;
 
             // Reference patterns are not loaded when we are in dump mode
@@ -32,6 +33,6 @@ class BayesF16:public Client::Suite
             arm_gaussian_naive_bayes_instance_f16 bayes;
 
             const float16_t *inp;
-            float16_t *bufp;
+            float16_t *bufp,*tempp;
 
     };

CMSIS/DSP/Testing/Include/Benchmarks/BayesF32.h

@@ -17,6 +17,7 @@ class BayesF32:public Client::Suite
             Client::Pattern<int16_t> dims;
 
             Client::LocalPattern<float32_t> outputProbas;
+            Client::LocalPattern<float32_t> temp;
             Client::LocalPattern<int16_t> outputPredicts;
 
             // Reference patterns are not loaded when we are in dump mode
@@ -32,6 +33,6 @@ class BayesF32:public Client::Suite
             arm_gaussian_naive_bayes_instance_f32 bayes;
 
             const float32_t *inp;
-            float32_t *bufp;
+            float32_t *bufp,*tempp;
 
     };

CMSIS/DSP/Testing/Include/Tests/BayesF16.h

@@ -17,6 +17,7 @@ class BayesF16:public Client::Suite
             Client::Pattern<int16_t> dims;
 
             Client::LocalPattern<float16_t> outputProbas;
+            Client::LocalPattern<float16_t> temp;
             Client::LocalPattern<int16_t> outputPredicts;
 
             // Reference patterns are not loaded when we are in dump mode

CMSIS/DSP/Testing/Include/Tests/BayesF32.h

@@ -17,6 +17,7 @@ class BayesF32:public Client::Suite
             Client::Pattern<int16_t> dims;
 
             Client::LocalPattern<float32_t> outputProbas;
+            Client::LocalPattern<float32_t> temp;
             Client::LocalPattern<int16_t> outputPredicts;
 
             // Reference patterns are not loaded when we are in dump mode

CMSIS/DSP/Testing/Source/Benchmarks/BayesF16.cpp

@@ -11,7 +11,7 @@
 
        p = arm_gaussian_naive_bayes_predict_f16(&bayes, 
                 inp, 
-                bufp);
+                bufp,tempp);
 
     } 
 
@@ -51,6 +51,7 @@
             predicts.reload(BayesF16::PREDICTS2_S16_ID,mgr);
 
             outputProbas.create(this->classNb,BayesF16::OUT_PROBA_F16_ID,mgr);
+            temp.create(this->classNb,BayesF16::OUT_PROBA_F16_ID,mgr);
 
             bayes.vectorDimension=this->vecDim;
             bayes.numberOfClasses=this->classNb;
@@ -62,6 +63,7 @@
             this->inp = input.ptr() + nbi;
 
             this->bufp = outputProbas.ptr();
+            this->tempp = temp.ptr();
 
           }
           break;

CMSIS/DSP/Testing/Source/Benchmarks/BayesF32.cpp

@@ -11,7 +11,7 @@
 
        p = arm_gaussian_naive_bayes_predict_f32(&bayes, 
                 inp, 
-                bufp);
+                bufp,tempp);
 
     } 
 
@@ -51,6 +51,7 @@
             predicts.reload(BayesF32::PREDICTS2_S16_ID,mgr);
 
             outputProbas.create(this->classNb,BayesF32::OUT_PROBA_F32_ID,mgr);
+            temp.create(this->classNb,BayesF32::OUT_PROBA_F32_ID,mgr);
 
             bayes.vectorDimension=this->vecDim;
             bayes.numberOfClasses=this->classNb;
@@ -62,6 +63,7 @@
             this->inp = input.ptr() + nbi;
 
             this->bufp = outputProbas.ptr();
+            this->tempp = temp.ptr();
 
           }
           break;

CMSIS/DSP/Testing/Source/Tests/BayesF16.cpp

@@ -10,6 +10,7 @@
        const float16_t *inp = input.ptr();
 
        float16_t *bufp = outputProbas.ptr();
+       float16_t *tempp = temp.ptr();
        int16_t *p = outputPredicts.ptr();
 
        
@@ -17,7 +18,7 @@
        {
           *p = arm_gaussian_naive_bayes_predict_f16(&bayes, 
                 inp, 
-                bufp);
+                bufp,tempp);
 
           inp += this->vecDim;
           bufp += this->classNb;
@@ -62,6 +63,7 @@
             predicts.reload(BayesF16::PREDICTS1_S16_ID,mgr);
 
             outputProbas.create(this->nbPatterns*this->classNb,BayesF16::OUT_PROBA_F16_ID,mgr);
+            temp.create(this->nbPatterns*this->classNb,BayesF16::OUT_PROBA_F16_ID,mgr);
             outputPredicts.create(this->nbPatterns,BayesF16::OUT_PREDICT_S16_ID,mgr);
 
             bayes.vectorDimension=this->vecDim;

CMSIS/DSP/Testing/Source/Tests/BayesF32.cpp

@@ -10,6 +10,7 @@
        const float32_t *inp = input.ptr();
 
        float32_t *bufp = outputProbas.ptr();
+       float32_t *tempp = temp.ptr();
        int16_t *p = outputPredicts.ptr();
 
        
@@ -17,7 +18,7 @@
        {
           *p = arm_gaussian_naive_bayes_predict_f32(&bayes, 
                 inp, 
-                bufp);
+                bufp,tempp);
 
           inp += this->vecDim;
           bufp += this->classNb;
@@ -62,6 +63,8 @@
             predicts.reload(BayesF32::PREDICTS1_S16_ID,mgr);
 
             outputProbas.create(this->nbPatterns*this->classNb,BayesF32::OUT_PROBA_F32_ID,mgr);
+            temp.create(this->nbPatterns*this->classNb,BayesF32::OUT_PROBA_F32_ID,mgr);
+
             outputPredicts.create(this->nbPatterns,BayesF32::OUT_PREDICT_S16_ID,mgr);
 
             bayes.vectorDimension=this->vecDim;