[hamradio-commits] [gnss-sdr] 72/149: fix sse implementation
Carles Fernandez
carles_fernandez-guest at moszumanska.debian.org
Sat Feb 6 19:43:04 UTC 2016
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 a2429a851cabdc6c3bef54831fc40c94411e831d
Author: Carles Fernandez <carles.fernandez at gmail.com>
Date: Sat Jan 16 22:52:10 2016 +0100
fix sse implementation
---
.../kernels/volk_gnsssdr/volk_gnsssdr_16ic_x2_dot_prod_16ic.h | 8 ++++----
.../kernels/volk_gnsssdr/volk_gnsssdr_16ic_x2_multiply_16ic.h | 3 +--
2 files changed, 5 insertions(+), 6 deletions(-)
diff --git a/src/algorithms/libs/volk_gnsssdr_module/volk_gnsssdr/kernels/volk_gnsssdr/volk_gnsssdr_16ic_x2_dot_prod_16ic.h b/src/algorithms/libs/volk_gnsssdr_module/volk_gnsssdr/kernels/volk_gnsssdr/volk_gnsssdr_16ic_x2_dot_prod_16ic.h
index 75be955..af102db 100644
--- a/src/algorithms/libs/volk_gnsssdr_module/volk_gnsssdr/kernels/volk_gnsssdr/volk_gnsssdr_16ic_x2_dot_prod_16ic.h
+++ b/src/algorithms/libs/volk_gnsssdr_module/volk_gnsssdr/kernels/volk_gnsssdr/volk_gnsssdr_16ic_x2_dot_prod_16ic.h
@@ -58,7 +58,7 @@ static inline void volk_gnsssdr_16ic_x2_dot_prod_16ic_generic(lv_16sc_t* result,
{
//r*a.r - i*a.i, i*a.r + r*a.i
lv_16sc_t tmp = in_a[n] * in_b[n];
- result[0] = lv_cmake(sat_adds16i(lv_creal(result[0]), lv_creal(tmp)), sat_adds16i(lv_cimag(result[0]), lv_cimag(tmp) ));
+ result[0] += lv_cmake(sat_adds16i(lv_creal(tmp), lv_creal(tmp)), sat_adds16i(lv_cimag(tmp), lv_cimag(tmp) ));
}
//printf("generic result = %i,%i", lv_creal(result[0]),lv_cimag(result[0]));
}
@@ -127,14 +127,14 @@ static inline void volk_gnsssdr_16ic_x2_dot_prod_16ic_a_sse2(lv_16sc_t* out, con
for (int i = 0; i < 4; ++i)
{
- dotProduct = lv_cmake(sat_adds16i(lv_creal(dotProduct), lv_creal(dotProductVector[i])), sat_adds16i(lv_cimag(dotProduct), lv_cimag(dotProductVector[i])));
+ dotProduct += lv_cmake(sat_adds16i(lv_creal(dotProductVector[i]), lv_creal(dotProductVector[i])), sat_adds16i(lv_cimag(dotProductVector[i]), lv_cimag(dotProductVector[i])));
}
}
- for (unsigned int i = 0; i < (num_points % 4); ++i)
+ for (unsigned int i = sse_iters * 4; i < num_points; ++i)
{
lv_16sc_t tmp = (*_in_a++) * (*_in_b++);
- dotProduct = lv_cmake( sat_adds16i(lv_creal(dotProduct), lv_creal(tmp)), sat_adds16i(lv_cimag(dotProduct), lv_cimag(tmp)));
+ dotProduct += lv_cmake( sat_adds16i(lv_creal(tmp), lv_creal(tmp)), sat_adds16i(lv_cimag(tmp), lv_cimag(tmp)));
}
*_out = dotProduct;
diff --git a/src/algorithms/libs/volk_gnsssdr_module/volk_gnsssdr/kernels/volk_gnsssdr/volk_gnsssdr_16ic_x2_multiply_16ic.h b/src/algorithms/libs/volk_gnsssdr_module/volk_gnsssdr/kernels/volk_gnsssdr/volk_gnsssdr_16ic_x2_multiply_16ic.h
index 1336824..d01b07c 100644
--- a/src/algorithms/libs/volk_gnsssdr_module/volk_gnsssdr/kernels/volk_gnsssdr/volk_gnsssdr_16ic_x2_multiply_16ic.h
+++ b/src/algorithms/libs/volk_gnsssdr_module/volk_gnsssdr/kernels/volk_gnsssdr/volk_gnsssdr_16ic_x2_multiply_16ic.h
@@ -74,7 +74,6 @@ static inline void volk_gnsssdr_16ic_x2_multiply_16ic_a_sse2(lv_16sc_t* out, con
lv_16sc_t* _out = out;
for(unsigned int number = 0; number < sse_iters; number++)
{
-
//std::complex<T> memory structure: real part -> reinterpret_cast<cv T*>(a)[2*i]
//imaginery part -> reinterpret_cast<cv T*>(a)[2*i + 1]
// a[127:0]=[a3.i,a3.r,a2.i,a2.r,a1.i,a1.r,a0.i,a0.r]
@@ -104,7 +103,7 @@ static inline void volk_gnsssdr_16ic_x2_multiply_16ic_a_sse2(lv_16sc_t* out, con
_out += 4;
}
- for (unsigned int i = 0; i < (num_points % 4); ++i)
+ for (unsigned int i = sse_iters * 4; i < num_points; ++i)
{
*_out++ = (*_in_a++) * (*_in_b++);
}
--
Alioth's /usr/local/bin/git-commit-notice on /srv/git.debian.org/git/pkg-hamradio/gnss-sdr.git
More information about the pkg-hamradio-commits
mailing list