Loosen up the accuracy bounds a tiny bit in some tests for NVIDIA cards, which have...

[movit] / effect_chain_test.cpp

diff --git a/effect_chain_test.cpp b/effect_chain_test.cpp
index bbb58a0587409d6cca03d867e110b34dee6c6a3c..97b90ed5d331386cbef60725725c8a5d9a49e09b 100644 (file)
--- a/effect_chain_test.cpp
+++ b/effect_chain_test.cpp
@@ -1367,8 +1367,8 @@ TEST(EffectChainTest, Linear10bitIntermediateAccuracy) {
        }
 
        // This maximum error is pretty bad; about 6.5 levels of a 10-bit sRGB
-       // framebuffer.
-       expect_equal(linear_data, out_data, size, 1, 7e-3, 2e-5);
+       // framebuffer. (Slightly more on NVIDIA cards.)
+       expect_equal(linear_data, out_data, size, 1, 7.5e-3, 2e-5);
 }
 
 TEST(EffectChainTest, SquareRoot10bitIntermediateAccuracy) {
@@ -1397,7 +1397,7 @@ TEST(EffectChainTest, SquareRoot10bitIntermediateAccuracy) {
        // This maximum error is much better; about 0.7 levels of a 10-bit sRGB
        // framebuffer (ideal would be 0.5). That is an order of magnitude better
        // than in the linear test above. The RMS error is much better, too.
-       expect_equal(linear_data, out_data, size, 1, 7e-4, 5e-6);
+       expect_equal(linear_data, out_data, size, 1, 7.5e-4, 5e-6);
 }
 
 TEST(EffectChainTest, SquareRootIntermediateIsTurnedOffForNonLinearData) {