NMSIS/DSP: benchmark fix SupportFunctions test

NMSIS/DSP/Benchmark/SupportFunctions/funcs_def.h

@@ -53,7 +53,7 @@ extern void q31ToQ15_riscv_q31_to_q15();
 extern void sort_riscv_sort_f32();
 extern void mergeSort_riscv_merge_sort_f32();
 
-extern void weightedSum_riscv_weighted_sum_f16();
-extern void weightedSum_riscv_weighted_sum_f32();
+extern void weightedSum_riscv_weighted_average_f16();
+extern void weightedSum_riscv_weighted_average_f32();
 
 #endif

NMSIS/DSP/Benchmark/SupportFunctions/riscv_dsp_benchmark.c

@@ -58,8 +58,8 @@ int main()
     sort_riscv_sort_f32();
     mergeSort_riscv_merge_sort_f32();
 
-    weightedSum_riscv_weighted_sum_f16();
-    weightedSum_riscv_weighted_sum_f32();
+    weightedSum_riscv_weighted_average_f16();
+    weightedSum_riscv_weighted_average_f32();
 
     printf("All tests are passed.\n");
     printf("test for SupportFunctions benchmark finished.\n");

NMSIS/DSP/Benchmark/SupportFunctions/test_riscv_weighted_sum_f16.c → NMSIS/DSP/Benchmark/SupportFunctions/test_riscv_weighted_average_f16.c

@@ -2,11 +2,11 @@
 #include <stdio.h>
 #include "validate.h"
 #include "nmsis_bench.h"
-#include "TestData/SupportFunctions/weighted_sum_f16/test_data.h"
+#include "TestData/SupportFunctions/weighted_average_f16/test_data.h"
 
 BENCH_DECLARE_VAR();
 
-void weightedSum_riscv_weighted_sum_f16(void)
+void weightedSum_riscv_weighted_average_f16(void)
 {
 #if defined (RISCV_FLOAT16_SUPPORTED)
     float16_t f16_weighted_output;
@@ -16,9 +16,9 @@ void weightedSum_riscv_weighted_sum_f16(void)
         f16_weighted_array[i] = (float16_t)rand() / RAND_MAX;
     }
 
-    BENCH_START(riscv_weighted_sum_f16);
-    f16_weighted_output = riscv_weighted_sum_f16(f16_weighted_sum_array, f16_weighted_array, WEIGHT_NUM);
-    BENCH_END(riscv_weighted_sum_f16);
+    BENCH_START(riscv_weighted_average_f16);
+    f16_weighted_output = riscv_weighted_average_f16(f16_weighted_sum_array, f16_weighted_array, WEIGHT_NUM);
+    BENCH_END(riscv_weighted_average_f16);
 
 #endif /* #if defined (RISCV_FLOAT16_SUPPORTED) */
 }

NMSIS/DSP/Benchmark/SupportFunctions/test_riscv_weighted_sum_f32.c → NMSIS/DSP/Benchmark/SupportFunctions/test_riscv_weighted_average_f32.c

@@ -2,11 +2,11 @@
 #include <stdio.h>
 #include "validate.h"
 #include "nmsis_bench.h"
-#include "TestData/SupportFunctions/weighted_sum_f32/test_data.h"
+#include "TestData/SupportFunctions/weighted_average_f32/test_data.h"
 
 BENCH_DECLARE_VAR();
 
-void weightedSum_riscv_weighted_sum_f32(void)
+void weightedSum_riscv_weighted_average_f32(void)
 {
     float32_t f32_weighted_output;
 
@@ -15,7 +15,7 @@ void weightedSum_riscv_weighted_sum_f32(void)
         f32_weighted_array[i] = (float32_t)rand() / RAND_MAX;
     }
 
-    BENCH_START(riscv_weighted_sum_f32);
-    f32_weighted_output = riscv_weighted_sum_f32(f32_weighted_sum_array, f32_weighted_array, WEIGHT_NUM);
-    BENCH_END(riscv_weighted_sum_f32);
+    BENCH_START(riscv_weighted_average_f32);
+    f32_weighted_output = riscv_weighted_average_f32(f32_weighted_sum_array, f32_weighted_array, WEIGHT_NUM);
+    BENCH_END(riscv_weighted_average_f32);
 }

NMSIS/DSP/Source/SupportFunctions/riscv_weighted_average_f16.c

@@ -77,14 +77,33 @@ RISCV_DSP_ATTRIBUTE float16_t riscv_weighted_average_f16(const float16_t *in, co
     accum1=0.0f16;
     accum2=0.0f16;
 
+#if defined(RISCV_MATH_VECTOR)
+    uint32_t blkCnt_v = blockSize;
+    size_t l;
+    vfloat16m8_t v_x, v_y;
+    vfloat16m1_t v_a, v_b;
+    l = __riscv_vsetvl_e16m1(1);
+    v_a = __riscv_vfsub_vv_f16m1(v_a, v_a, l);
+    v_b = __riscv_vfsub_vv_f16m1(v_b, v_b, l);
+    for (; (l = __riscv_vsetvl_e16m8(blkCnt_v)) > 0; blkCnt_v -= l) {
+        v_x = __riscv_vle16_v_f16m8(pIn, l);
+        pIn += l;
+        v_y = __riscv_vle16_v_f16m8(pW, l);
+        pW += l;
+        v_a = __riscv_vfredusum_vs_f16m8_f16m1(__riscv_vfmul_vv_f16m8(v_x, v_y, l), v_a, l);
+        v_b = __riscv_vfredusum_vs_f16m8_f16m1(v_y, v_b, l);
+    }
+    accum1 += __riscv_vfmv_f_s_f16m1_f16(v_a);
+    accum2 += __riscv_vfmv_f_s_f16m1_f16(v_b);
+#else
     blkCnt = blockSize;
     while(blkCnt > 0)
     {
         accum1 += (_Float16)*pIn++ * (_Float16)*pW;
         accum2 += (_Float16)*pW++;
         blkCnt--;
     }
-
+#endif /* #if defined(RISCV_MATH_VECTOR) */
     return(accum1 / accum2);
 }
 /**

NMSIS/DSP/Source/SupportFunctions/riscv_weighted_average_f32.c

@@ -63,14 +63,33 @@ RISCV_DSP_ATTRIBUTE float32_t riscv_weighted_average_f32(const float32_t *in, co
     accum1=0.0f;
     accum2=0.0f;
 
+#if defined(RISCV_MATH_VECTOR)
+    uint32_t blkCnt_v = blockSize;
+    size_t l;
+    vfloat32m8_t v_x, v_y;
+    vfloat32m1_t v_a, v_b;
+    l = __riscv_vsetvl_e32m1(1);
+    v_a = __riscv_vfsub_vv_f32m1(v_a, v_a, l);
+    v_b = __riscv_vfsub_vv_f32m1(v_b, v_b, l);
+    for (; (l = __riscv_vsetvl_e32m8(blkCnt_v)) > 0; blkCnt_v -= l) {
+        v_x = __riscv_vle32_v_f32m8(pIn, l);
+        pIn += l;
+        v_y = __riscv_vle32_v_f32m8(pW, l);
+        pW += l;
+        v_a = __riscv_vfredusum_vs_f32m8_f32m1(__riscv_vfmul_vv_f32m8(v_x, v_y, l), v_a, l);
+        v_b = __riscv_vfredusum_vs_f32m8_f32m1(v_y, v_b, l);
+    }
+    accum1 += __riscv_vfmv_f_s_f32m1_f32(v_a);
+    accum2 += __riscv_vfmv_f_s_f32m1_f32(v_b);
+#else
     blkCnt = blockSize;
     while(blkCnt > 0)
     {
         accum1 += *pIn++ * *pW;
         accum2 += *pW++;
         blkCnt--;
     }
-
+#endif /* #if defined(RISCV_MATH_VECTOR) */
     return(accum1 / accum2);
 }