add n-r for rvv target

Signed-off-by: Magnus Lundmark <[email protected]>

Author: Ka-zam

kernels/volk/volk_32f_invsqrt_32f.h

@@ -310,18 +310,16 @@ volk_32f_invsqrt_32f_rvv(float* cVector, const float* aVector, unsigned int num_
     for (size_t vl; n > 0; n -= vl, aVector += vl, cVector += vl) {
         vl = __riscv_vsetvl_e32m8(n);
         vfloat32m8_t a = __riscv_vle32_v_f32m8(aVector, vl);
+        vfloat32m8_t half = __riscv_vfmv_v_f_f32m8(0.5f, vl);
+        vfloat32m8_t three_halfs = __riscv_vfmv_v_f_f32m8(1.5f, vl);
         // Initial estimate (~7-bit precision)
         vfloat32m8_t x = __riscv_vfrsqrt7(a, vl);
         // Two Newton-Raphson iterations: x = x * (1.5 - 0.5 * a * x * x)
-        vfloat32m8_t ax = __riscv_vfmul(a, x, vl);
-        vfloat32m8_t half_ax = __riscv_vfmul_vf(ax, 0.5f, vl);
+        vfloat32m8_t half_a = __riscv_vfmul(half, a, vl);
         x = __riscv_vfmul(
-            x, __riscv_vfnmsac_vf(__riscv_vfmv_v_f_f32m8(1.5f, vl), half_ax, x, vl), vl);
-        // Second iteration
-        ax = __riscv_vfmul(a, x, vl);
-        half_ax = __riscv_vfmul_vf(ax, 0.5f, vl);
+            x, __riscv_vfnmsac(three_halfs, half_a, __riscv_vfmul(x, x, vl), vl), vl);
         x = __riscv_vfmul(
-            x, __riscv_vfnmsac_vf(__riscv_vfmv_v_f_f32m8(1.5f, vl), half_ax, x, vl), vl);
+            x, __riscv_vfnmsac(three_halfs, half_a, __riscv_vfmul(x, x, vl), vl), vl);
         __riscv_vse32(cVector, x, vl);
     }
 }