summaryrefslogtreecommitdiff
path: root/volk/lib/qa_32u_byteswap_aligned16.cc
diff options
context:
space:
mode:
authorEric Blossom2010-12-21 13:31:58 -0800
committerEric Blossom2010-12-21 13:31:58 -0800
commit75b650ef2506fe5c607517a3d5188b0705f60fde (patch)
tree99cb6fd6adc3e91edc832cbaee236d2fe30fa3fd /volk/lib/qa_32u_byteswap_aligned16.cc
parent1984aab345857296564e174395cf3af1375e1469 (diff)
parented78ba5d9999bbe50507373a1aa2877ef0da64c6 (diff)
downloadgnuradio-75b650ef2506fe5c607517a3d5188b0705f60fde.tar.gz
gnuradio-75b650ef2506fe5c607517a3d5188b0705f60fde.tar.bz2
gnuradio-75b650ef2506fe5c607517a3d5188b0705f60fde.zip
Merge branch 'next' into guile.
Passes distcheck. * next: (32 commits) volk: Fix for popcnt's 64/32-bit issues. Using a copy of config.guess and config.sub instead of sym links. Including time header to qa files. Changed python env variable to more globally usable version. gr_uhd: Quick fix for make distcheck failures if UHD is not installed. PFB resampler: fix it this way to avoid the signed/unsigned warning. PFB resampler: fixes bug where filter could be looking past the number of inputs. resampler PFB: Sets relative rate when rate is changed. Create method to set rate on pfb_arb_resamp after it has been created. Allow it to be called from GRC. volk: fix for running 32-bit OS on 64-bit processor. System is correctly identified as 32-bit and compiles with the correct flags. volk: changing the path variables again. This works on my various systems tested. Using abs_ path names failed on Ubuntu 8.04 32-bit. volk: May be a hack, but it was required for my 32-bit Fedora 13 to work. volk: Removing unnecessary shell script; last commit takes care if its functions. volk: Fixing build system to handle making volk_mktables, volk_tables.h, and volk_config.h instead of a standalone shell script. volk: readding 16sc_magnitude_32f_sse with fix for SSE hadd_ps error. volk: Adding a few more generic-only test cases. volk: adding generic QA test for 16sc_magnitude_32f. volk: modified the configure scripts to output which architectures it will be building based on the configure tests. uhd: update notes in grc blocks for addressing scheme volk: Removing erroneous SSE function that actually usese an SSE3 intrin (mm_hadd_ps). ...
Diffstat (limited to 'volk/lib/qa_32u_byteswap_aligned16.cc')
-rw-r--r--volk/lib/qa_32u_byteswap_aligned16.cc60
1 files changed, 60 insertions, 0 deletions
diff --git a/volk/lib/qa_32u_byteswap_aligned16.cc b/volk/lib/qa_32u_byteswap_aligned16.cc
new file mode 100644
index 000000000..313c786b6
--- /dev/null
+++ b/volk/lib/qa_32u_byteswap_aligned16.cc
@@ -0,0 +1,60 @@
+#include <volk/volk.h>
+#include <qa_32u_byteswap_aligned16.h>
+#include <volk/volk_32u_byteswap_aligned16.h>
+#include <cstdlib>
+#include <cstring>
+#include <ctime>
+
+//test for sse
+
+#ifndef LV_HAVE_SSE2
+
+void qa_32u_byteswap_aligned16::t1() {
+ printf("sse2 not available... no test performed\n");
+}
+
+#else
+
+void qa_32u_byteswap_aligned16::t1() {
+
+ volk_environment_init();
+ clock_t start, end;
+ double total;
+ const int vlen = 3201;
+ const int ITERS = 100001;
+
+ uint32_t output0[vlen] __attribute__ ((aligned (16)));
+ uint32_t output01[vlen] __attribute__ ((aligned (16)));
+
+ for(int i = 0; i < vlen; ++i) {
+ output0[i] = (uint32_t) ((rand() - (RAND_MAX/2)) / (RAND_MAX/2));
+ }
+ memcpy(output01, output0, vlen*sizeof(uint32_t));
+ printf("32u_byteswap_aligned\n");
+
+ start = clock();
+ for(int count = 0; count < ITERS; ++count) {
+ volk_32u_byteswap_aligned16_manual(output0, vlen, "generic");
+ }
+ end = clock();
+ total = (double)(end-start)/(double)CLOCKS_PER_SEC;
+ printf("generic_time: %f\n", total);
+ start = clock();
+ for(int count = 0; count < ITERS; ++count) {
+ volk_32u_byteswap_aligned16_manual(output01, vlen, "sse2");
+ }
+ end = clock();
+ total = (double)(end-start)/(double)CLOCKS_PER_SEC;
+ printf("sse2_time: %f\n", total);
+ for(int i = 0; i < 1; ++i) {
+ //printf("inputs: %d, %d\n", input0[i*2], input0[i*2 + 1]);
+ //printf("generic... %d, ssse3... %d\n", output0[i], output1[i]);
+ }
+
+ for(int i = 0; i < vlen; ++i) {
+ //printf("%d...%d\n", output0[i], output01[i]);
+ CPPUNIT_ASSERT_EQUAL(output0[i], output01[i]);
+ }
+}
+
+#endif