[hamradio-commits] [gnss-sdr] 210/236: Adding AVX protokernel

[Carles Fernandez]

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

carles_fernandez-guest pushed a commit to branch next
in repository gnss-sdr.

commit b2cc18bfa56004c157b023d05f66401929005082
Author: Carles Fernandez <carles.fernandez at gmail.com>
Date:   Sat Apr 23 16:31:41 2016 +0200

    Adding AVX protokernel
---
 .../volk_gnsssdr/volk_gnsssdr_16ic_convert_32fc.h  | 58 ++++++++++++++++++++--
 1 file changed, 53 insertions(+), 5 deletions(-)

diff --git a/src/algorithms/libs/volk_gnsssdr_module/volk_gnsssdr/kernels/volk_gnsssdr/volk_gnsssdr_16ic_convert_32fc.h b/src/algorithms/libs/volk_gnsssdr_module/volk_gnsssdr/kernels/volk_gnsssdr/volk_gnsssdr_16ic_convert_32fc.h
index 4873cc8..23f125c 100644
--- a/src/algorithms/libs/volk_gnsssdr_module/volk_gnsssdr/kernels/volk_gnsssdr/volk_gnsssdr_16ic_convert_32fc.h
+++ b/src/algorithms/libs/volk_gnsssdr_module/volk_gnsssdr/kernels/volk_gnsssdr/volk_gnsssdr_16ic_convert_32fc.h
@@ -96,7 +96,6 @@ static inline void volk_gnsssdr_16ic_convert_32fc_a_sse2(lv_32fc_t* outputVector
             *_out++ = lv_cmake((float)lv_creal(*_in), (float)lv_cimag(*_in));
             _in++;
         }
-
 }
 #endif /* LV_HAVE_SSE2 */
 
@@ -123,13 +122,62 @@ static inline void volk_gnsssdr_16ic_convert_32fc_u_sse2(lv_32fc_t* outputVector
             *_out++ = lv_cmake((float)lv_creal(*_in), (float)lv_cimag(*_in));
             _in++;
         }
-
 }
 #endif /* LV_HAVE_SSE2 */
 
-// SSE4.1
-// a = _mm_load_si128((__m128i*)_in); //load (2 byte imag, 2 byte real) x 4 into 128 bits reg
-//use _mm_cvtepi16_epi32 !!!!
+
+#ifdef LV_HAVE_AVX
+#include <immintrin.h>
+
+static inline void volk_gnsssdr_16ic_convert_32fc_u_axv(lv_32fc_t* outputVector, const lv_16sc_t* inputVector, unsigned int num_points)
+{
+    const unsigned int sse_iters = num_points / 4;
+
+    const lv_16sc_t* _in = inputVector;
+    lv_32fc_t* _out = outputVector;
+    __m256 a;
+    for(unsigned int number = 0; number < sse_iters; number++)
+        {
+            a = _mm256_set_ps((float)(lv_cimag(_in[3])), (float)(lv_creal(_in[3])), (float)(lv_cimag(_in[2])), (float)(lv_creal(_in[2])), (float)(lv_cimag(_in[1])), (float)(lv_creal(_in[1])), (float)(lv_cimag(_in[0])), (float)(lv_creal(_in[0]))); // //load (2 byte imag, 2 byte real) x 2 into 128 bits reg
+            _mm256_storeu_ps((float*)_out, a);
+            _in += 4;
+            _out += 4;
+        }
+    _mm256_zeroupper();
+    for (unsigned int i = 0; i < (num_points % 4); ++i)
+        {
+            *_out++ = lv_cmake((float)lv_creal(*_in), (float)lv_cimag(*_in));
+            _in++;
+        }
+}
+#endif /* LV_HAVE_AVX */
+
+#ifdef LV_HAVE_AVX
+#include <immintrin.h>
+
+static inline void volk_gnsssdr_16ic_convert_32fc_a_axv(lv_32fc_t* outputVector, const lv_16sc_t* inputVector, unsigned int num_points)
+{
+    const unsigned int sse_iters = num_points / 4;
+
+    const lv_16sc_t* _in = inputVector;
+    lv_32fc_t* _out = outputVector;
+    __m256 a;
+    for(unsigned int number = 0; number < sse_iters; number++)
+        {
+            a = _mm256_set_ps((float)(lv_cimag(_in[3])), (float)(lv_creal(_in[3])), (float)(lv_cimag(_in[2])), (float)(lv_creal(_in[2])), (float)(lv_cimag(_in[1])), (float)(lv_creal(_in[1])), (float)(lv_cimag(_in[0])), (float)(lv_creal(_in[0]))); // //load (2 byte imag, 2 byte real) x 2 into 128 bits reg
+            _mm256_store_ps((float*)_out, a);
+            _in += 4;
+            _out += 4;
+        }
+    _mm256_zeroupper();
+    for (unsigned int i = 0; i < (num_points % 4); ++i)
+        {
+            *_out++ = lv_cmake((float)lv_creal(*_in), (float)lv_cimag(*_in));
+            _in++;
+        }
+}
+#endif /* LV_HAVE_AVX */
+
 
 #ifdef LV_HAVE_NEON
 #include <arm_neon.h>

-- 
Alioth's /usr/local/bin/git-commit-notice on /srv/git.debian.org/git/pkg-hamradio/gnss-sdr.git