Solved bugs and compilation error in simd and vector_simd

lib/include/srslte/phy/utils/simd.h

@@ -226,7 +226,7 @@ static inline simd_f_t srslte_simd_f_mul(simd_f_t a, simd_f_t b) {
 static inline simd_f_t srslte_simd_f_addsub(simd_f_t a, simd_f_t b) {
 #ifdef LV_HAVE_AVX512
   __m512 r = _mm512_add_ps(a, b);
-  return _mm512_mask_sub_ps(r, 0b1010101010101010, a, b);
+  return _mm512_mask_sub_ps(r, 0b0101010101010101, a, b);
 #else /* LV_HAVE_AVX512 */
 #ifdef LV_HAVE_AVX2
   return _mm256_addsub_ps(a, b);
@@ -642,10 +642,10 @@ static inline simd_s_t srslte_simd_s_load(int16_t *ptr) {
   return _mm512_load_si512(ptr);
 #else /* LV_HAVE_AVX512 */
 #ifdef LV_HAVE_AVX2
-  return _mm256_load_si256(ptr);
+  return _mm256_load_si256((__m256i*) ptr);
 #else /* LV_HAVE_AVX2 */
 #ifdef LV_HAVE_SSE
-  return _mm_load_si128(ptr);
+  return _mm_load_si128((__m128i*) ptr);
 #endif /* LV_HAVE_SSE */
 #endif /* LV_HAVE_AVX2 */
 #endif /* LV_HAVE_AVX512 */
@@ -653,13 +653,13 @@ static inline simd_s_t srslte_simd_s_load(int16_t *ptr) {
 
 static inline simd_s_t srslte_simd_s_loadu(int16_t *ptr) {
 #ifdef LV_HAVE_AVX512
-  return _mm512_load_si512(ptr);
+  return _mm512_loadu_si512(ptr);
 #else /* LV_HAVE_AVX512 */
   #ifdef LV_HAVE_AVX2
-  return _mm256_load_si256(ptr);
+  return _mm256_loadu_si256((__m256i*) ptr);
 #else /* LV_HAVE_AVX2 */
 #ifdef LV_HAVE_SSE
-  return _mm_load_si128(ptr);
+  return _mm_loadu_si128((__m128i*) ptr);
 #endif /* LV_HAVE_SSE */
 #endif /* LV_HAVE_AVX2 */
 #endif /* LV_HAVE_AVX512 */
@@ -670,10 +670,10 @@ static inline void srslte_simd_s_store(int16_t *ptr, simd_s_t simdreg) {
   _mm512_store_si512(ptr, simdreg);
 #else /* LV_HAVE_AVX512 */
 #ifdef LV_HAVE_AVX2
-  _mm256_store_si256(ptr, simdreg);
+  _mm256_store_si256((__m256i*) ptr, simdreg);
 #else /* LV_HAVE_AVX2 */
 #ifdef LV_HAVE_SSE
-  _mm_store_si128(ptr, simdreg);
+  _mm_store_si128((__m128i*) ptr, simdreg);
 #endif /* LV_HAVE_SSE */
 #endif /* LV_HAVE_AVX2 */
 #endif /* LV_HAVE_AVX512 */
@@ -684,10 +684,10 @@ static inline void srslte_simd_s_storeu(int16_t *ptr, simd_s_t simdreg) {
   _mm512_storeu_si512(ptr, simdreg);
 #else /* LV_HAVE_AVX512 */
 #ifdef LV_HAVE_AVX2
-  _mm256_storeu_si256(ptr, simdreg);
+  _mm256_storeu_si256((__m256i*) ptr, simdreg);
 #else /* LV_HAVE_AVX2 */
 #ifdef LV_HAVE_SSE
-  _mm_storeu_si128(ptr, simdreg);
+  _mm_storeu_si128((__m128i*) ptr, simdreg);
 #endif /* LV_HAVE_SSE */
 #endif /* LV_HAVE_AVX2 */
 #endif /* LV_HAVE_AVX512 */

lib/include/srslte/phy/utils/vector_simd.h

@@ -44,7 +44,7 @@ extern "C" {
 #ifdef LV_HAVE_SSE
 #define SRSLTE_IS_ALIGNED(PTR) (((size_t)(PTR) & 0x0F) == 0)
 #else /* LV_HAVE_SSE */
-#define SRSLTE_IS_ALIGNED(PTR) (true)
+#define SRSLTE_IS_ALIGNED(PTR) (1)
 #endif /* LV_HAVE_SSE */
 #endif /* LV_HAVE_AVX */
 #endif /* LV_HAVE_AVX512 */

lib/src/phy/utils/test/vector_test.c

@@ -45,7 +45,7 @@ bool mmse_solver = false;
 bool verbose = false;
 
 #define MAX_MSE (1e-3)
-#define NOF_REPETITIONS (1024*128)
+#define NOF_REPETITIONS (1024)
 #define MAX_FUNCTIONS (64)
 #define MAX_BLOCKS (16)
 
@@ -70,7 +70,7 @@ bool verbose = false;
     return passed;\
 }
 
-#define MALLOC(TYPE, NAME) TYPE *NAME = srslte_vec_malloc(sizeof(TYPE)*block_size)
+#define MALLOC(TYPE, NAME) TYPE *NAME = malloc(sizeof(TYPE)*block_size)
 
 
 static double elapsed_us(struct timeval *ts_start, struct timeval *ts_end) {
@@ -339,7 +339,7 @@ TEST(srslte_vec_prod_conj_ccc,
 TEST(srslte_vec_sc_prod_ccc,
   MALLOC(cf_t, x);
   MALLOC(cf_t, z);
-  cf_t y = RANDOM_F();
+  cf_t y = RANDOM_CF();
 
   cf_t gold;
   for (int i = 0; i < block_size; i++) {
@@ -469,7 +469,7 @@ int main(int argc, char **argv) {
   uint32_t func_count = 0;
   bool passed = true;
 
-  for (uint32_t block_size = 1; block_size <= 1024*16; block_size *= 2) {
+  for (uint32_t block_size = 1; block_size <= 1024*8; block_size *= 2) {
     func_count = 0;
 
     passed &= test_srslte_vec_dot_prod_sss(func_names[func_count], &timmings[func_count][size_count], block_size);

lib/src/phy/utils/vector_simd.c

@@ -77,7 +77,7 @@ int srslte_vec_dot_prod_sss_simd(int16_t *x, int16_t *y, int len) {
 
 void srslte_vec_sum_sss_simd(int16_t *x, int16_t *y, int16_t *z, int len) {
   int i = 0;
-#ifdef SRSLTE_SIMD_S_SIZE
+#if SRSLTE_SIMD_S_SIZE
   if (SRSLTE_IS_ALIGNED(x) && SRSLTE_IS_ALIGNED(y) && SRSLTE_IS_ALIGNED(z)) {
     for (; i < len - SRSLTE_SIMD_S_SIZE + 1; i += SRSLTE_SIMD_S_SIZE) {
       simd_s_t a = srslte_simd_s_load(&x[i]);
@@ -106,7 +106,7 @@ void srslte_vec_sum_sss_simd(int16_t *x, int16_t *y, int16_t *z, int len) {
 
 void srslte_vec_sub_sss_simd(int16_t *x, int16_t *y, int16_t *z, int len) {
   int i = 0;
-#ifdef SRSLTE_SIMD_S_SIZE
+#if SRSLTE_SIMD_S_SIZE
   if (SRSLTE_IS_ALIGNED(x) && SRSLTE_IS_ALIGNED(y) && SRSLTE_IS_ALIGNED(z)) {
     for (; i < len - SRSLTE_SIMD_S_SIZE + 1; i += SRSLTE_SIMD_S_SIZE) {
       simd_s_t a = srslte_simd_s_load(&x[i]);
@@ -135,7 +135,7 @@ void srslte_vec_sub_sss_simd(int16_t *x, int16_t *y, int16_t *z, int len) {
 
 void srslte_vec_prod_sss_simd(int16_t *x, int16_t *y, int16_t *z, int len) {
   int i = 0;
-#ifdef SRSLTE_SIMD_S_SIZE
+#if SRSLTE_SIMD_S_SIZE
   if (SRSLTE_IS_ALIGNED(x) && SRSLTE_IS_ALIGNED(y) && SRSLTE_IS_ALIGNED(z)) {
     for (; i < len - SRSLTE_SIMD_S_SIZE + 1; i += SRSLTE_SIMD_S_SIZE) {
       simd_s_t a = srslte_simd_s_load(&x[i]);
@@ -721,14 +721,14 @@ void srslte_vec_sc_prod_ccc_simd(cf_t *x, cf_t h, cf_t *z, int len) {
     }
   } else {
     for (; i < len - SRSLTE_SIMD_F_SIZE / 2 + 1; i += SRSLTE_SIMD_F_SIZE / 2) {
-      simd_f_t temp = srslte_simd_f_load((float *) &x[i]);
+      simd_f_t temp = srslte_simd_f_loadu((float *) &x[i]);
 
       simd_f_t m1 = srslte_simd_f_mul(hre, temp);
       simd_f_t sw = srslte_simd_f_swap(temp);
       simd_f_t m2 = srslte_simd_f_mul(him, sw);
       simd_f_t r = srslte_simd_f_addsub(m1, m2);
 
-      srslte_simd_f_store((float *) &z[i], r);
+      srslte_simd_f_storeu((float *) &z[i], r);
     }
   }
 #endif