From: Steinar H. Gunderson <sgunderson@bigfoot.com>
Date: Sun, 26 Feb 2017 15:15:10 +0000 (+0100)
Subject: Loosen up the accuracy bounds a tiny bit in some tests for NVIDIA cards, which have... 
X-Git-Tag: 1.5.0~8
X-Git-Url: https://git.sesse.net/?p=movit;a=commitdiff_plain;h=1ca8a02bbbea840ab4165bf1d14a588d71db9e0c

Loosen up the accuracy bounds a tiny bit in some tests for NVIDIA cards, which have worse rounding (we do not bother to add manual rounding for these cards, since the difference is so small).
---

diff --git a/effect_chain_test.cpp b/effect_chain_test.cpp
index bbb58a0..97b90ed 100644
--- 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) {