[Commit-gnuradio] [gnuradio] 03/07: volk: removing unused variables in neon

[git]

This is an automated email from the git hooks/post-receive script.

trondeau pushed a commit to branch master
in repository gnuradio.

commit 17b9ba03d6501c82a553b8e1f910258461320539
Author: Nathan West <address@hidden>
Date:   Fri Jul 25 18:07:51 2014 -0400

    volk: removing unused variables in neon
---
 volk/kernels/volk/volk_16i_max_star_16i.h       |  2 +-
 volk/kernels/volk/volk_32f_x2_dot_prod_32f.h    |  2 +-
 volk/kernels/volk/volk_32f_x3_sum_of_poly_32f.h | 11 ++---------
 volk/kernels/volk/volk_32fc_32f_dot_prod_32fc.h |  7 -------
 volk/kernels/volk/volk_32fc_conjugate_32fc.h    |  5 -----
 volk/kernels/volk/volk_32fc_x2_dot_prod_32fc.h  |  5 ++---
 volk/kernels/volk/volk_32fc_x2_multiply_32fc.h  |  6 ++----
 volk/kernels/volk/volk_8i_convert_16i.h         |  1 -
 8 files changed, 8 insertions(+), 31 deletions(-)

diff --git a/volk/kernels/volk/volk_16i_max_star_16i.h 
b/volk/kernels/volk/volk_16i_max_star_16i.h
index 5366a2e..d296c43 100644
--- a/volk/kernels/volk/volk_16i_max_star_16i.h
+++ b/volk/kernels/volk/volk_16i_max_star_16i.h
@@ -91,7 +91,7 @@ static inline void volk_16i_max_star_16i_neon(short* target, 
short* src0, unsign
     const unsigned int eighth_points = num_points / 8;
     unsigned number;
     int16x8_t input_vec;
-    int16x8_t diff, max_vec, zeros;
+    int16x8_t diff, zeros;
     uint16x8_t comp1, comp2;
     zeros = veorq_s16(zeros, zeros);
     
diff --git a/volk/kernels/volk/volk_32f_x2_dot_prod_32f.h 
b/volk/kernels/volk/volk_32f_x2_dot_prod_32f.h
index ed16d9a..e8fa8b5 100644
--- a/volk/kernels/volk/volk_32f_x2_dot_prod_32f.h
+++ b/volk/kernels/volk/volk_32f_x2_dot_prod_32f.h
@@ -586,7 +586,7 @@ static inline void volk_32f_x2_dot_prod_32f_neonopts(float 
* result, const float
     const float* bPtr=  taps;
     unsigned int number = 0;
 
-    float32x4x4_t a_val, b_val, accumulator0, accumulator1;
+    float32x4x4_t a_val, b_val, accumulator0;
     accumulator0.val[0] = vdupq_n_f32(0);
     accumulator0.val[1] = vdupq_n_f32(0);
     accumulator0.val[2] = vdupq_n_f32(0);
diff --git a/volk/kernels/volk/volk_32f_x3_sum_of_poly_32f.h 
b/volk/kernels/volk/volk_32f_x3_sum_of_poly_32f.h
index c555bbb..3a5c710 100644
--- a/volk/kernels/volk/volk_32f_x3_sum_of_poly_32f.h
+++ b/volk/kernels/volk/volk_32f_x3_sum_of_poly_32f.h
@@ -305,15 +305,9 @@ static inline void 
volk_32f_x3_sum_of_poly_32f_a_neon(float* __restrict target,
   float32x2_t cutoff_vector;
   float32x2x2_t x_low, x_high;
   float32x4_t x_qvector, c_qvector, cpa_qvector;
-  float accumulator, final_result;
+  float accumulator;
   float res_accumulators[4];
   
-  float dbg_cpa[4], dbg_x[4], dbg_c[4];
-  float dbg_max[4];
-  float dbg_x_to_1[2], dbg_x_to_2[2], dbg_x_to_3[2], dbg_x_to_4[2];
-  float dbg_x_high[2], dbg_x_low[2];
-  float dbg_foo;
-
   c_qvector = vld1q_f32( zero );
   // load the cutoff in to a vector
   cutoff_vector = vdup_n_f32( *cutoff );
@@ -357,7 +351,7 @@ static inline void 
volk_32f_x3_sum_of_poly_32f_neonvert(float* __restrict target
   int i;
   float zero[4] = {0.0f, 0.0f, 0.0f, 0.0f };
 
-  float accumulator, final_result;
+  float accumulator;
   
 
   float32x4_t accumulator1_vec, accumulator2_vec, accumulator3_vec, 
accumulator4_vec;
@@ -407,7 +401,6 @@ static inline void 
volk_32f_x3_sum_of_poly_32f_neonvert(float* __restrict target
   accumulator = res_accumulators[0] + res_accumulators[1] + 
           res_accumulators[2] + res_accumulators[3];
 
-  float result = 0.0;
   float fst = 0.0;
   float sq = 0.0;
   float thrd = 0.0;
diff --git a/volk/kernels/volk/volk_32fc_32f_dot_prod_32fc.h 
b/volk/kernels/volk/volk_32fc_32f_dot_prod_32fc.h
index cf67c13..6d0c848 100644
--- a/volk/kernels/volk/volk_32fc_32f_dot_prod_32fc.h
+++ b/volk/kernels/volk/volk_32fc_32f_dot_prod_32fc.h
@@ -296,8 +296,6 @@ static inline void volk_32fc_32f_dot_prod_32fc_neon_unroll 
( lv_32fc_t* __restri
    const float* inputPtr = (float*)input;
    const float* tapsPtr = taps;
    float zero[4] = {0.0f, 0.0f, 0.0f, 0.0f };
-   float* real_accum;
-   float current_accum = 0.0f ;
    float accVector_real[4]; 
    float accVector_imag[4];
 
@@ -307,7 +305,6 @@ static inline void volk_32fc_32f_dot_prod_32fc_neon_unroll 
( lv_32fc_t* __restri
    float32x4_t  tmp_real1, tmp_imag1;
    float32x4_t real_accumulator0, imag_accumulator0;
    float32x4_t real_accumulator1, imag_accumulator1;
-
   
    // zero out accumulators
    // take a *float, return float32x4_t
@@ -315,7 +312,6 @@ static inline void volk_32fc_32f_dot_prod_32fc_neon_unroll 
( lv_32fc_t* __restri
    imag_accumulator0 = vld1q_f32( zero );
    real_accumulator1 = vld1q_f32( zero );
    imag_accumulator1 = vld1q_f32( zero );
-   float dbgVec[8];
  
    for(number=0 ;number < quarterPoints; number++){
       // load doublewords and duplicate in to second lane
@@ -379,8 +375,6 @@ static inline void volk_32fc_32f_dot_prod_32fc_a_neon ( 
lv_32fc_t* __restrict re
    const float* inputPtr = (float*)input;
    const float* tapsPtr = taps;
    float zero[4] = {0.0f, 0.0f, 0.0f, 0.0f };
-   float* real_accum;
-   float current_accum = 0.0f ;
    float accVector_real[4];
    float accVector_imag[4];
 
@@ -415,7 +409,6 @@ static inline void volk_32fc_32f_dot_prod_32fc_a_neon ( 
lv_32fc_t* __restrict re
 
    }
 
-   // void vst1q_f32( float32_t * ptr, float32x4_t val);
    // store results back to a complex (array of 2 floats)
    vst1q_f32(accVector_real, real_accumulator);
    vst1q_f32(accVector_imag, imag_accumulator);
diff --git a/volk/kernels/volk/volk_32fc_conjugate_32fc.h 
b/volk/kernels/volk/volk_32fc_conjugate_32fc.h
index 480fa36..0f74b01 100644
--- a/volk/kernels/volk/volk_32fc_conjugate_32fc.h
+++ b/volk/kernels/volk/volk_32fc_conjugate_32fc.h
@@ -122,11 +122,6 @@ static inline void 
volk_32fc_conjugate_32fc_a_neon(lv_32fc_t* cVector, const lv_
     lv_32fc_t* c = cVector;
     const lv_32fc_t* a = aVector;
 
-    float conj[4] = {-0.f, -0.f, -0.f, -0.f};
-    //uint32x4_t conjugator;
-
-    //conjugator = vld1q_u32( (uint32_t *)conj );
-
     for(number=0; number < quarterPoints; number++){
       __builtin_prefetch(a+4);
       x = vld2q_f32((float*)a); // Load the complex data as ar,br,cr,dr; 
ai,bi,ci,di
diff --git a/volk/kernels/volk/volk_32fc_x2_dot_prod_32fc.h 
b/volk/kernels/volk/volk_32fc_x2_dot_prod_32fc.h
index fb79d66..5301c35 100644
--- a/volk/kernels/volk/volk_32fc_x2_dot_prod_32fc.h
+++ b/volk/kernels/volk/volk_32fc_x2_dot_prod_32fc.h
@@ -826,8 +826,8 @@ static inline void 
volk_32fc_x2_dot_prod_32fc_neon_opttests(lv_32fc_t* result, c
     lv_32fc_t* b_ptr = (lv_32fc_t*) input;
     // for 2-lane vectors, 1st lane holds the real part,
     // 2nd lane holds the imaginary part
-    float32x4x2_t a_val, b_val, c_val, accumulator;
-    float32x4x2_t tmp_real, tmp_imag;
+    float32x4x2_t a_val, b_val, accumulator;
+    float32x4x2_t tmp_imag;
     accumulator.val[0] = vdupq_n_f32(0);
     accumulator.val[1] = vdupq_n_f32(0);
     
@@ -875,7 +875,6 @@ static inline void 
volk_32fc_x2_dot_prod_32fc_neon_optfma(lv_32fc_t* result, con
     // for 2-lane vectors, 1st lane holds the real part,
     // 2nd lane holds the imaginary part
     float32x4x2_t a_val, b_val, accumulator1, accumulator2;
-    float32x4x2_t tmp_real, tmp_imag;
     accumulator1.val[0] = vdupq_n_f32(0);
     accumulator1.val[1] = vdupq_n_f32(0);
     accumulator2.val[0] = vdupq_n_f32(0);
diff --git a/volk/kernels/volk/volk_32fc_x2_multiply_32fc.h 
b/volk/kernels/volk/volk_32fc_x2_multiply_32fc.h
index 0993a16..8d2d48b 100644
--- a/volk/kernels/volk/volk_32fc_x2_multiply_32fc.h
+++ b/volk/kernels/volk/volk_32fc_x2_multiply_32fc.h
@@ -214,12 +214,10 @@ static inline void 
volk_32fc_x2_multiply_32fc_neon_opttests(lv_32fc_t* cVector,
     lv_32fc_t *a_ptr = (lv_32fc_t*) aVector;
     lv_32fc_t *b_ptr = (lv_32fc_t*) bVector;
     unsigned int quarter_points = num_points / 4;
-    float32x4x2_t a_val, b_val, c_val;
-    float32x4x2_t tmp_real, tmp_imag;
+    float32x4x2_t a_val, b_val;
+    float32x4x2_t tmp_imag;
     unsigned int number = 0;
 
-    // TODO: I suspect the compiler is doing a poor job scheduling this. This 
seems
-    // highly optimal, but is barely better than generic
     for(number = 0; number < quarter_points; ++number) {
         a_val = vld2q_f32((float*)a_ptr); // a0r|a1r|a2r|a3r || a0i|a1i|a2i|a3i
         b_val = vld2q_f32((float*)b_ptr); // b0r|b1r|b2r|b3r || b0i|b1i|b2i|b3i
diff --git a/volk/kernels/volk/volk_8i_convert_16i.h 
b/volk/kernels/volk/volk_8i_convert_16i.h
index 5b27900..3b89a3f 100644
--- a/volk/kernels/volk/volk_8i_convert_16i.h
+++ b/volk/kernels/volk/volk_8i_convert_16i.h
@@ -151,7 +151,6 @@ static inline void volk_8i_convert_16i_neon(int16_t* 
outputVector, const int8_t*
     const int8_t* inputVectorPtr = inputVector;
     unsigned int number;
     const unsigned int eighth_points = num_points / 8;
-    float scale_factor = 256;
 
     int8x8_t input_vec ;
     int16x8_t converted_vec;