mirrors/gnss-sdr
1
0
Fork 0
mirror of https://github.com/gnss-sdr/gnss-sdr synced 2025-01-16 04:05:46 +00:00
Code Issues Releases Wiki Activity

Revert "prefetching data in the cache"

Browse Source 
This reverts commit aac79eb78a.
This commit is contained in:
Javier Arribas 2016-03-09 15:44:55 +01:00
parent d664dc63b3
commit b03e3fc3af
2 changed files with 6 additions and 6 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

5
src/algorithms/libs/volk_gnsssdr_module/volk_gnsssdr/kernels/volk_gnsssdr/volk_gnsssdr_16ic_s32fc_x2_rotator_16ic.h
View File

@ -103,7 +103,6 @@ static inline void volk_gnsssdr_16ic_s32fc_x2_rotator_16ic_a_sse3(lv_16sc_t* out
for(unsigned int number = 0; number < sse_iters; number++) for(unsigned int number = 0; number < sse_iters; number++)
{ {
a = _mm_set_ps((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 a = _mm_set_ps((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
__builtin_prefetch(_in + 8);
//complex 32fc multiplication b=a*two_phase_acc_reg //complex 32fc multiplication b=a*two_phase_acc_reg
yl = _mm_moveldup_ps(two_phase_acc_reg); // Load yl with cr,cr,dr,dr yl = _mm_moveldup_ps(two_phase_acc_reg); // Load yl with cr,cr,dr,dr
yh = _mm_movehdup_ps(two_phase_acc_reg); // Load yh with ci,ci,di,di yh = _mm_movehdup_ps(two_phase_acc_reg); // Load yh with ci,ci,di,di
@ -150,7 +149,7 @@ static inline void volk_gnsssdr_16ic_s32fc_x2_rotator_16ic_a_sse3(lv_16sc_t* out
_out += 4; _out += 4;
} }
_mm_store_ps((float*)two_phase_acc, two_phase_acc_reg); _mm_storeu_ps((float*)two_phase_acc, two_phase_acc_reg);
(*phase) = two_phase_acc[0]; (*phase) = two_phase_acc[0];
for (unsigned int i = sse_iters * 4; i < num_points; ++i) for (unsigned int i = sse_iters * 4; i < num_points; ++i)
@ -200,7 +199,6 @@ static inline void volk_gnsssdr_16ic_s32fc_x2_rotator_16ic_u_sse3(lv_16sc_t* out
for(unsigned int number = 0; number < sse_iters; number++) for(unsigned int number = 0; number < sse_iters; number++)
{ {
a = _mm_set_ps((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 a = _mm_set_ps((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
__builtin_prefetch(_in + 8);
//complex 32fc multiplication b=a*two_phase_acc_reg //complex 32fc multiplication b=a*two_phase_acc_reg
yl = _mm_moveldup_ps(two_phase_acc_reg); // Load yl with cr,cr,dr,dr yl = _mm_moveldup_ps(two_phase_acc_reg); // Load yl with cr,cr,dr,dr
yh = _mm_movehdup_ps(two_phase_acc_reg); // Load yh with ci,ci,di,di yh = _mm_movehdup_ps(two_phase_acc_reg); // Load yh with ci,ci,di,di
@ -221,6 +219,7 @@ static inline void volk_gnsssdr_16ic_s32fc_x2_rotator_16ic_u_sse3(lv_16sc_t* out
//next two samples //next two samples
_in += 2; _in += 2;
a = _mm_set_ps((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 a = _mm_set_ps((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
__builtin_prefetch(_in + 8);
//complex 32fc multiplication b=a*two_phase_acc_reg //complex 32fc multiplication b=a*two_phase_acc_reg
yl = _mm_moveldup_ps(two_phase_acc_reg); // Load yl with cr,cr,dr,dr yl = _mm_moveldup_ps(two_phase_acc_reg); // Load yl with cr,cr,dr,dr
yh = _mm_movehdup_ps(two_phase_acc_reg); // Load yh with ci,ci,di,di yh = _mm_movehdup_ps(two_phase_acc_reg); // Load yh with ci,ci,di,di

7
src/algorithms/libs/volk_gnsssdr_module/volk_gnsssdr/kernels/volk_gnsssdr/volk_gnsssdr_16ic_x2_rotator_dot_prod_16ic_xn.h
View File

@ -140,7 +140,6 @@ static inline void volk_gnsssdr_16ic_x2_rotator_dot_prod_16ic_xn_a_sse3(lv_16sc_
// Phase rotation on operand in_common starts here: // Phase rotation on operand in_common starts here:
//printf("generic phase %i: %f,%f\n", n*4,lv_creal(*phase),lv_cimag(*phase)); //printf("generic phase %i: %f,%f\n", n*4,lv_creal(*phase),lv_cimag(*phase));
pa = _mm_set_ps((float)(lv_cimag(_in_common[1])), (float)(lv_creal(_in_common[1])), (float)(lv_cimag(_in_common[0])), (float)(lv_creal(_in_common[0]))); // //load (2 byte imag, 2 byte real) x 2 into 128 bits reg pa = _mm_set_ps((float)(lv_cimag(_in_common[1])), (float)(lv_creal(_in_common[1])), (float)(lv_cimag(_in_common[0])), (float)(lv_creal(_in_common[0]))); // //load (2 byte imag, 2 byte real) x 2 into 128 bits reg
__builtin_prefetch(_in_common + 8);
//complex 32fc multiplication b=a*two_phase_acc_reg //complex 32fc multiplication b=a*two_phase_acc_reg
yl = _mm_moveldup_ps(two_phase_acc_reg); // Load yl with cr,cr,dr,dr yl = _mm_moveldup_ps(two_phase_acc_reg); // Load yl with cr,cr,dr,dr
yh = _mm_movehdup_ps(two_phase_acc_reg); // Load yh with ci,ci,di,di yh = _mm_movehdup_ps(two_phase_acc_reg); // Load yh with ci,ci,di,di
@ -304,6 +303,7 @@ static inline void volk_gnsssdr_16ic_x2_rotator_dot_prod_16ic_xn_u_sse3(lv_16sc_
for(unsigned int number = 0; number < sse_iters; number++) for(unsigned int number = 0; number < sse_iters; number++)
{ {
// Phase rotation on operand in_common starts here: // Phase rotation on operand in_common starts here:
pa = _mm_set_ps((float)(lv_cimag(_in_common[1])), (float)(lv_creal(_in_common[1])), (float)(lv_cimag(_in_common[0])), (float)(lv_creal(_in_common[0]))); // //load (2 byte imag, 2 byte real) x 2 into 128 bits reg pa = _mm_set_ps((float)(lv_cimag(_in_common[1])), (float)(lv_creal(_in_common[1])), (float)(lv_cimag(_in_common[0])), (float)(lv_creal(_in_common[0]))); // //load (2 byte imag, 2 byte real) x 2 into 128 bits reg
__builtin_prefetch(_in_common + 8); __builtin_prefetch(_in_common + 8);
//complex 32fc multiplication b=a*two_phase_acc_reg //complex 32fc multiplication b=a*two_phase_acc_reg
@ -378,7 +378,7 @@ static inline void volk_gnsssdr_16ic_x2_rotator_dot_prod_16ic_xn_u_sse3(lv_16sc_
a = _mm_or_si128(realcacc[n_vec], imagcacc[n_vec]); a = _mm_or_si128(realcacc[n_vec], imagcacc[n_vec]);
_mm_store_si128((__m128i*)dotProductVector, a); // Store the results back into the dot product vector _mm_storeu_si128((__m128i*)dotProductVector, a); // Store the results back into the dot product vector
dotProduct = lv_cmake(0,0); dotProduct = lv_cmake(0,0);
for (int i = 0; i < 4; ++i) for (int i = 0; i < 4; ++i)
{ {
@ -390,9 +390,10 @@ static inline void volk_gnsssdr_16ic_x2_rotator_dot_prod_16ic_xn_u_sse3(lv_16sc_
free(realcacc); free(realcacc);
free(imagcacc); free(imagcacc);
_mm_store_ps((float*)two_phase_acc, two_phase_acc_reg); _mm_storeu_ps((float*)two_phase_acc, two_phase_acc_reg);
(*phase) = two_phase_acc[0]; (*phase) = two_phase_acc[0];
for(unsigned int n = sse_iters * 4; n < num_points; n++) for(unsigned int n = sse_iters * 4; n < num_points; n++)
{ {
tmp16 = in_common[n]; tmp16 = in_common[n];