From 5b45b875ed58fd66234764a05da42c6eaff22c4d Mon Sep 17 00:00:00 2001 From: Nick Foster Date: Tue, 11 Jan 2011 15:17:55 -0800 Subject: Volk: Added more Orc routines (including complex multiply). Started redoing the testing framework so it's easier to add new archs to tests. --- volk/lib/qa_32f_normalize_aligned16.cc | 13 +++++++++++++ 1 file changed, 13 insertions(+) (limited to 'volk/lib/qa_32f_normalize_aligned16.cc') diff --git a/volk/lib/qa_32f_normalize_aligned16.cc b/volk/lib/qa_32f_normalize_aligned16.cc index 1c7b485a6..0da43ecff 100644 --- a/volk/lib/qa_32f_normalize_aligned16.cc +++ b/volk/lib/qa_32f_normalize_aligned16.cc @@ -26,13 +26,16 @@ void qa_32f_normalize_aligned16::t1() { float* output0; float* output01; + float* output02; ret = posix_memalign((void**)&output0, 16, vlen*sizeof(float)); ret = posix_memalign((void**)&output01, 16, vlen*sizeof(float)); + ret = posix_memalign((void**)&output02, 16, vlen*sizeof(float)); for(int i = 0; i < vlen; ++i) { output0[i] = ((float) (rand() - (RAND_MAX/2))) / static_cast((RAND_MAX/2)); } memcpy(output01, output0, vlen*sizeof(float)); + memcpy(output02, output0, vlen*sizeof(float)); printf("32f_normalize_aligned\n"); start = clock(); @@ -49,6 +52,14 @@ void qa_32f_normalize_aligned16::t1() { end = clock(); total = (double)(end-start)/(double)CLOCKS_PER_SEC; printf("sse_time: %f\n", total); + start = clock(); + for(int count = 0; count < ITERS; ++count) { + volk_32f_normalize_aligned16_manual(output02, 1.15, vlen, "orc"); + } + end = clock(); + total = (double)(end-start)/(double)CLOCKS_PER_SEC; + printf("orc_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]); @@ -57,10 +68,12 @@ void qa_32f_normalize_aligned16::t1() { for(int i = 0; i < vlen; ++i) { // printf("%e...%e\n", output0[i], output01[i]); CPPUNIT_ASSERT_DOUBLES_EQUAL(output0[i], output01[i], fabs(output0[i])*1e-4); + CPPUNIT_ASSERT_DOUBLES_EQUAL(output0[i], output02[i], fabs(output0[i])*1e-4); } free(output0); free(output01); + free(output02); } #endif -- cgit From 6091bad60cdfdf21624da452c7a8b74405345070 Mon Sep 17 00:00:00 2001 From: Nick Foster Date: Fri, 21 Jan 2011 15:41:30 -0800 Subject: Volk: removed all the old QA code that is covered by the test framework. --- volk/lib/qa_32f_normalize_aligned16.cc | 79 ---------------------------------- 1 file changed, 79 deletions(-) delete mode 100644 volk/lib/qa_32f_normalize_aligned16.cc (limited to 'volk/lib/qa_32f_normalize_aligned16.cc') diff --git a/volk/lib/qa_32f_normalize_aligned16.cc b/volk/lib/qa_32f_normalize_aligned16.cc deleted file mode 100644 index 0da43ecff..000000000 --- a/volk/lib/qa_32f_normalize_aligned16.cc +++ /dev/null @@ -1,79 +0,0 @@ -#include -#include -#include -#include -#include -#include - -//test for sse - -#ifndef LV_HAVE_SSE - -void qa_32f_normalize_aligned16::t1() { - printf("sse not available... no test performed\n"); -} - -#else - -void qa_32f_normalize_aligned16::t1() { - - volk_environment_init(); - int ret; - clock_t start, end; - double total; - const int vlen = 320001; - const int ITERS = 100; - - float* output0; - float* output01; - float* output02; - ret = posix_memalign((void**)&output0, 16, vlen*sizeof(float)); - ret = posix_memalign((void**)&output01, 16, vlen*sizeof(float)); - ret = posix_memalign((void**)&output02, 16, vlen*sizeof(float)); - - for(int i = 0; i < vlen; ++i) { - output0[i] = ((float) (rand() - (RAND_MAX/2))) / static_cast((RAND_MAX/2)); - } - memcpy(output01, output0, vlen*sizeof(float)); - memcpy(output02, output0, vlen*sizeof(float)); - printf("32f_normalize_aligned\n"); - - start = clock(); - for(int count = 0; count < ITERS; ++count) { - volk_32f_normalize_aligned16_manual(output0, 1.15, 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_32f_normalize_aligned16_manual(output01, 1.15, vlen, "sse"); - } - end = clock(); - total = (double)(end-start)/(double)CLOCKS_PER_SEC; - printf("sse_time: %f\n", total); - start = clock(); - for(int count = 0; count < ITERS; ++count) { - volk_32f_normalize_aligned16_manual(output02, 1.15, vlen, "orc"); - } - end = clock(); - total = (double)(end-start)/(double)CLOCKS_PER_SEC; - printf("orc_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("%e...%e\n", output0[i], output01[i]); - CPPUNIT_ASSERT_DOUBLES_EQUAL(output0[i], output01[i], fabs(output0[i])*1e-4); - CPPUNIT_ASSERT_DOUBLES_EQUAL(output0[i], output02[i], fabs(output0[i])*1e-4); - } - - free(output0); - free(output01); - free(output02); -} - -#endif -- cgit