[PATCH v2 3/5] libipa: histogram: Fix quantile() calculation for fractional results
Stefan Klug
stefan.klug at ideasonboard.com
Tue Apr 1 14:36:11 CEST 2025
The calculation of the frac variable is based solely on integers and
therefore results in the fractional part being either 0 or 1.
In the original code from RaspberryPi this is mitigated by casting the
nominator to a double. This works for most cases, but fails when q is
very small because of the quantization introduced by item being an
integer.
Fix both issues by doing the full calculation in double and remove the
should_fail tag.
Signed-off-by: Stefan Klug <stefan.klug at ideasonboard.com>
Reviewed-by: Kieran Bingham <kieran.bingham at ideasonboard.com>
Reviewed-by: Laurent Pinchart <laurent.pinchart at ideasonboard.com>
---
Changes in v2:
- Order fix patch after test
- Remove should_fail tag
- Break line before operator instead of after
---
src/ipa/libipa/histogram.cpp | 3 ++-
test/ipa/libipa/meson.build | 2 +-
2 files changed, 3 insertions(+), 2 deletions(-)
diff --git a/src/ipa/libipa/histogram.cpp b/src/ipa/libipa/histogram.cpp
index 10e44b54a0cf..ea042f0a17b9 100644
--- a/src/ipa/libipa/histogram.cpp
+++ b/src/ipa/libipa/histogram.cpp
@@ -130,7 +130,8 @@ double Histogram::quantile(double q, uint32_t first, uint32_t last) const
if (cumulative_[first + 1] == cumulative_[first])
frac = 0;
else
- frac = (item - cumulative_[first]) / (cumulative_[first + 1] - cumulative_[first]);
+ frac = (q * total() - cumulative_[first])
+ / (cumulative_[first + 1] - cumulative_[first]);
return first + frac;
}
diff --git a/test/ipa/libipa/meson.build b/test/ipa/libipa/meson.build
index 83c84bd8c227..8c63ebd8e2f7 100644
--- a/test/ipa/libipa/meson.build
+++ b/test/ipa/libipa/meson.build
@@ -2,7 +2,7 @@
libipa_test = [
{'name': 'fixedpoint', 'sources': ['fixedpoint.cpp']},
- {'name': 'histogram', 'sources': ['histogram.cpp'], 'should_fail': true},
+ {'name': 'histogram', 'sources': ['histogram.cpp']},
{'name': 'interpolator', 'sources': ['interpolator.cpp']},
]
--
2.43.0
More information about the libcamera-devel
mailing list