From: Steinar H. Gunderson Date: Sat, 25 Jan 2014 18:43:29 +0000 (+0100) Subject: Make the ResampleEffect accuracy test stricter. X-Git-Tag: 1.0~52 X-Git-Url: https://git.sesse.net/?a=commitdiff_plain;h=9224f01a362117deef6f1140c6802fde256fb168;p=movit Make the ResampleEffect accuracy test stricter. First, make sure we test one individual pass, and that we test it in fp32. Second, set a limit that's actually grounded in something real, not just a pretty power of 10. --- diff --git a/resample_effect_test.cpp b/resample_effect_test.cpp index e4c23b1..dee08e3 100644 --- a/resample_effect_test.cpp +++ b/resample_effect_test.cpp @@ -168,22 +168,25 @@ TEST(ResampleEffectTest, UpscaleByThreeGetsCorrectPixelCenters) { } TEST(ResampleEffectTest, HeavyResampleGetsSumRight) { - const int swidth = 1280, sheight = 720; - const int dwidth = 36, dheight = 20; + // Do only one resample pass, more specifically the last one, which goes to + // our fp32 output. This allows us to analyze the precision without intermediate + // fp16 rounding. + const int swidth = 1, sheight = 1280; + const int dwidth = 1, dheight = 64; float data[swidth * sheight], out_data[dwidth * dheight], expected_data[dwidth * dheight]; for (int y = 0; y < sheight; ++y) { for (int x = 0; x < swidth; ++x) { - data[y * swidth + x] = 0.5f; + data[y * swidth + x] = 1.0f; } } for (int y = 0; y < dheight; ++y) { for (int x = 0; x < dwidth; ++x) { - expected_data[y * dwidth + x] = 0.5f; + expected_data[y * dwidth + x] = 1.0f; } } - EffectChainTester tester(NULL, dwidth, dheight, FORMAT_GRAYSCALE, COLORSPACE_sRGB, GAMMA_LINEAR); + EffectChainTester tester(NULL, dwidth, dheight, FORMAT_GRAYSCALE, COLORSPACE_sRGB, GAMMA_LINEAR, GL_RGBA32F); ImageFormat format; format.color_space = COLORSPACE_sRGB; @@ -198,6 +201,10 @@ TEST(ResampleEffectTest, HeavyResampleGetsSumRight) { ASSERT_TRUE(resample_effect->set_int("height", dheight)); tester.run(out_data, GL_RED, COLORSPACE_sRGB, GAMMA_LINEAR); - expect_equal(expected_data, out_data, dwidth, dheight, 0.001); + // Require that we are within 10-bit accuracy. Note that this is for + // one pass only; some cards that don't have correct fp32 -> fp16 + // rounding in the intermediate framebuffers will go outside this after + // a 2D resize. This limit is tight enough that it will be good enough + // for 8-bit accuracy, though. + expect_equal(expected_data, out_data, dwidth, dheight, 0.5 / 1023.0); } -