Something went wrong. Try again.
What happens if I push all of nixpkgs
Something went wrong. Try again.
3.7 kB · 108 lines
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109From fb8e3c74d582038a358936d827f53c4c0c43d4e6 Mon Sep 17 00:00:00 2001From: Matt Harvey <mattharvey@google.com>Date: Mon, 27 Nov 2023 16:28:53 -0800Subject: [PATCH] Fix testresample.c output span; add exit code
Prior to this chance, the "Resample with different factors" test onlypassed for 60 of the 63 factors, with the 3 failing ones being thelargest.
1. Since only 63 distinct factors were being considered, 100 random samples was overkill.2. To support noticing failure in continuous build systems, it's nice if the test exit()s with nonzero when there are failures.3. The root cause was a formula error when determining which indices in the resampled output ought be compared. Details are explained in a comment.--- tests/testresample.c | 32 ++++++++++++++++++++++++-------- 1 file changed, 24 insertions(+), 8 deletions(-)
diff --git a/tests/testresample.c b/tests/testresample.cindex aa83a46..640df5a 100644--- a/tests/testresample.c+++ b/tests/testresample.c@@ -19,6 +19,8 @@ #define MIN(A, B) (A) < (B)? (A) : (B) +int global_error;+ void runtest(int srclen, double freq, double factor, int srcblocksize, int dstblocksize) {@@ -65,10 +67,12 @@ void runtest(int srclen, double freq, double factor, if (o < 0) { printf("Error: resample_process returned an error: %d\n", o);+ global_error = 1; } if (out <= 0) { printf("Error: resample_process returned %d samples\n", out);+ global_error = 1; free(src); free(dst); return;@@ -79,15 +83,16 @@ void runtest(int srclen, double freq, double factor, printf(" Expected ~%d, got %d samples out\n", expectedlen, out); }- + sum = 0.0; sumsq = 0.0; errcount = 0.0; - /* Don't compute statistics on all output values; the last few- are guaranteed to be off because it's based on far less- interpolation. */- statlen = out - fwidth;+ /* Don't compute statistics on all output values; the last small fraction+ are guaranteed to be off since they are interpolated based on far fewer+ values. When upsampling, the length of the range where this concern+ applies is in direct proportion to the upsampling factor. */+ statlen = out - ((int)round(fwidth * factor)); for(i=0; i<statlen; i++) { double diff = sin((i/freq)/factor) - dst[i];@@ -117,6 +122,7 @@ void runtest(int srclen, double freq, double factor, printf(" i=%d: expected %.3f, got %.3f\n", i, sin((i/freq)/factor), dst[i]); printf(" At least %d samples had significant error.\n", errcount);+ global_error = 1; } err = sum / statlen; rmserr = sqrt(sumsq / statlen);@@ -130,6 +136,8 @@ int main(int argc, char **argv) int i, srclen, dstlen, ifreq; double factor; + global_error = 0;+ printf("\n*** Vary source block size*** \n\n"); srclen = 10000; ifreq = 100;@@ -172,11 +180,19 @@ int main(int argc, char **argv) printf("\n*** Resample with different factors ***\n\n"); srclen = 10000; ifreq = 100;- for(i=0; i<100; i++) {- factor = ((rand() % 64) + 1) / 4.0;+ for (i = 1; i < 64; i++) {+ factor = i / 4.0;+ dstlen = (int)(srclen * factor + 10);+ runtest(srclen, (double)ifreq, factor, srclen, dstlen);+ }++ printf("\n*** Resample with large factors ***\n\n");+ srclen = 200;+ ifreq = 100;+ for (factor = 25.0; factor < 1000.0; factor *= 1.7) { dstlen = (int)(srclen * factor + 10); runtest(srclen, (double)ifreq, factor, srclen, dstlen); } - return 0;+ return global_error; }