[Piglit] [PATCH] Add test to verify srgb conformance
Roland Scheidegger
rscheidegger_lists at hispeed.ch
Thu Aug 1 17:57:41 PDT 2013
Am 01.08.2013 18:22, schrieb Jose Fonseca:
> ----- Original Message -----
>> From: Roland Scheidegger <sroland at vmware.com>
>>
>> Tests accuracy of srgb->linear and linear->srgb conversion,
>> according to d3d10 rules as I understand them:
>> 1) srgb->linear is permitted a tolerance of 0.5 on the srgb side
>> (meaning the result converted back to srgb (but before float->int
>> conversion) using perfect formula must not deviate more than 0.5)
>> 2) linear->srgb is permitted a tolerance of 0.6 compared to using
>> perfect formula
>> 3) additionally all srgb values must stay the same when doing
>> srgb to linear and then linear to srgb conversion.
>> This test does not actually verify 2) (which would need exhaustive
>> test of all float values) so if some floats outside these generated
>> by srgb->linear conversion get mapped to arbitrary values that will
>> go unnoticed. Likewise, correct behavior for floats outside 0.0-1.0
>> is not verified (including INFs and NaNs - the former should get clamped
>> to 0/255 in the end, NaNs should also get mapped to 0).
>
> Looks great to me. Thanks.
>
> Minor suggestion inline.
>
>> ---
>> tests/general/CMakeLists.gl.txt | 1 +
>> tests/general/srgb_conformance.c | 238
>> ++++++++++++++++++++++++++++++++++++++
>> 2 files changed, 239 insertions(+)
>> create mode 100644 tests/general/srgb_conformance.c
>>
>> diff --git a/tests/general/CMakeLists.gl.txt
>> b/tests/general/CMakeLists.gl.txt
>> index 1933b13..601dafc 100644
>> --- a/tests/general/CMakeLists.gl.txt
>> +++ b/tests/general/CMakeLists.gl.txt
>> @@ -145,5 +145,6 @@ piglit_add_executable (quad-invariance quad-invariance.c)
>> piglit_add_executable (clear-accum clear-accum.c)
>> piglit_add_executable (vs-point_size-zero vs-point_size-zero.c)
>> piglit_add_executable (triangle-guardband-viewport
>> triangle-guardband-viewport.c)
>> +piglit_add_executable (srgb_conformance srgb_conformance.c)
>>
>> # vim: ft=cmake:
>> diff --git a/tests/general/srgb_conformance.c
>> b/tests/general/srgb_conformance.c
>> new file mode 100644
>> index 0000000..dbf224e
>> --- /dev/null
>> +++ b/tests/general/srgb_conformance.c
>> @@ -0,0 +1,238 @@
>> +/*
>> + * Copyright © 2013 VMware, Inc.
>> + *
>> + * Permission is hereby granted, free of charge, to any person obtaining a
>> + * copy of this software and associated documentation files (the
>> "Software"),
>> + * to deal in the Software without restriction, including without limitation
>> + * the rights to use, copy, modify, merge, publish, distribute, sublicense,
>> + * and/or sell copies of the Software, and to permit persons to whom the
>> + * Software is furnished to do so, subject to the following conditions:
>> + *
>> + * The above copyright notice and this permission notice (including the next
>> + * paragraph) shall be included in all copies or substantial portions of the
>> + * Software.
>> + *
>> + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
>> OR
>> + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
>> + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
>> + * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR
>> OTHER
>> + * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
>> + * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
>> DEALINGS
>> + * IN THE SOFTWARE.
>> + *
>> + * Authors:
>> + * Roland Scheidegger <sroland at vmware.com>
>> + *
>> + * Tests accuracy of srgb->linear and linear->srgb conversion,
>> + * according to d3d10 rules:
>> + * 1) srgb->linear is permitted a tolerance of 0.5 on the srgb side
>> + * (meaning the result converted back to srgb (but before float->int
>> + * conversion) using perfect formula must not deviate more than 0.5)
>> + * 2) linear->srgb is permitted a tolerance of 0.6 compared to using
>> + * perfect formula
>> + * 3) additionally all srgb values must stay the same when doing
>> + * srgb to linear and then linear to srgb conversion.
>> + * This test does not actually verify 2) (which would need exhaustive
>> + * test of all float values) so if some floats outside these generated
>> + * by srgb->linear conversion get mapped to arbitrary values that will
>> + * go unnoticed. Likewise, correct behavior for floats outside 0.0-1.0
>> + * is not verified (including INFs and NaNs - the former should get clamped
>> + * to 0/255 in the end, NaNs should also get mapped to 0).
>> + *
>> + */
>> +
>> +#include "piglit-util-gl-common.h"
>> +
>> +PIGLIT_GL_TEST_CONFIG_BEGIN
>> +
>> + config.supports_gl_compat_version = 10;
>> +
>> + config.window_width = 16;
>> + config.window_height = 16;
>> + config.window_visual = PIGLIT_GL_VISUAL_RGBA | PIGLIT_GL_VISUAL_DOUBLE;
>> +
>> +PIGLIT_GL_TEST_CONFIG_END
>> +
>> +
>> +static GLfloat
>> +linear_to_nonlinear(float x)
>> +{
>> + if (x >= 1.0f)
>> + return 255.0f;
>> + else if (x >= 0.0031308f)
>> + return 255.0f * (1.055f * powf(x, 0.41666f) - 0.055f);
>> + else if (x > 0.0f)
>> + return 255.0f * (12.92f * x);
>> + else
>> + return 0.0f;
>> +}
>> +
>> +static enum piglit_result test_format(void)
>> +{
>> + GLboolean pass = GL_TRUE;
>> + GLuint texsrgb, texfb, fb;
>> + GLenum status;
>> + int i;
>> + float tex_vals[256][4];
>> + float readf[256][4];
>> + float tolerance;
>> + GLubyte readb[256][4];
>> + float maxErr = 0.0f;
>> +
>> + for (i = 0; i < 256; i++) {
>> + tex_vals[i][0] = (float)i / 255.0f;
>> + tex_vals[i][1] = tex_vals[i][2] = 0.0f;
>> + tex_vals[i][3] = 1.0f;
>> + }
>> +
>> + /* initialize texture */
>> + glGenTextures(1, &texsrgb);
>> + glBindTexture(GL_TEXTURE_2D, texsrgb);
>> + glTexImage2D(GL_TEXTURE_2D, 0, GL_SRGB8_ALPHA8_EXT,
>> + 16, 16, 0,
>> + GL_RGBA, GL_FLOAT, &tex_vals[0][0]);
>> +
>> + glGenFramebuffersEXT(1, &fb);
>> + glBindFramebufferEXT(GL_FRAMEBUFFER_EXT, fb);
>> + glViewport(0, 0, piglit_width, piglit_height);
>> +
>> + /* draw into float framebuffer and verify results */
>> + glGenTextures(1, &texfb);
>> + glBindTexture(GL_TEXTURE_2D, texfb);
>> + glTexImage2D(GL_TEXTURE_2D, 0, GL_RGBA32F_ARB,
>> + piglit_width, piglit_height, 0,
>> + GL_RGBA, GL_FLOAT, NULL);
>> +
>> + glFramebufferTexture2DEXT(GL_FRAMEBUFFER_EXT,
>> + GL_COLOR_ATTACHMENT0_EXT,
>> + GL_TEXTURE_2D,
>> + texfb,
>> + 0);
>> + assert(glGetError() == 0);
>> +
>> + status = glCheckFramebufferStatusEXT(GL_FRAMEBUFFER_EXT);
>> + printf("Testing using fb float format");
>> + if (status != GL_FRAMEBUFFER_COMPLETE_EXT) {
>> + printf(" - fbo incomplete (status = %s)\n",
>> + piglit_get_gl_enum_name(status));
>> + piglit_report_subtest_result(PIGLIT_SKIP, "float fb");
>> + return PIGLIT_SKIP;
>> + }
>> + printf("\n");
>> +
>> + glColor4f(1, 1, 1, 1);
>> + glEnable(GL_TEXTURE_2D);
>> + glBindTexture(GL_TEXTURE_2D, texsrgb);
>> +
>> + glTexEnvi(GL_TEXTURE_ENV, GL_TEXTURE_ENV_MODE, GL_COMBINE);
>> + glTexEnvi(GL_TEXTURE_ENV, GL_COMBINE_RGB, GL_REPLACE);
>> + glTexEnvi(GL_TEXTURE_ENV, GL_COMBINE_ALPHA, GL_REPLACE);
>> +
>> + glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_NEAREST);
>> + glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MAG_FILTER, GL_NEAREST);
>> +
>> + piglit_draw_rect_tex(-1, -1, 2, 2,
>> + 0, 0, 1, 1);
>> +
>> + glDisable(GL_TEXTURE_2D);
>> +
>> + /* have to make up our own error measuring, since we
>> + measure error on srgb side (so with values mapped back to srgb
>> + using accurate formula) */
>> +
>> + tolerance = 0.5f; /* as required by d3d10 */
>> + glReadPixels(0, 0, 16, 16, GL_RGBA, GL_FLOAT, &readf[0][0]);
>> +
>> + for (i = 0; i < 256; i++) {
>> +// printf("readback: %f observed: %f expected: %f\n", readf[i][0],
>> +// linear_to_nonlinear(readf[i][0]), (float)i);
>
> If this printfs could be handy in the future, I'd prefer to have them inside "if (0) { ...} ": it's less work to enable, and guarantees that the code does rot and is propely highlighted in editors, etc.
Ok. They are somewhat interesting for software implementation, though I
suspect
most hw has much lower error so the numbers don't tell much.
They were very much needed for debugging the test app though :-).
>> + float err = fabs(linear_to_nonlinear(readf[i][0]) - (float)i);
>> + if (err > maxErr) {
>> + maxErr = err;
>> + }
>> + if (err > tolerance) {
>> + printf(" failed when testing srgb->float result\n");
>> + pass = GL_FALSE;
>> + break;
>> + }
>> + }
>> + printf("max error srgb->linear was %f\n", maxErr);
>> +
>> + piglit_present_results();
>> +
>> + piglit_report_subtest_result(pass ? PIGLIT_PASS : PIGLIT_FAIL,
>> + "srgb->linear");
>> +
>> + /* draw into srgb framebuffer and verify results */
>> + glBindTexture(GL_TEXTURE_2D, texfb);
>> + glTexImage2D(GL_TEXTURE_2D, 0, GL_SRGB8_ALPHA8_EXT,
>> + piglit_width, piglit_height, 0,
>> + GL_RGBA, GL_FLOAT, NULL);
>> +
>> + glFramebufferTexture2DEXT(GL_FRAMEBUFFER_EXT,
>> + GL_COLOR_ATTACHMENT0_EXT,
>> + GL_TEXTURE_2D,
>> + texfb,
>> + 0);
>> +
>> + assert(glGetError() == 0);
>> +
>> + glEnable(GL_FRAMEBUFFER_SRGB_EXT);
>> + status = glCheckFramebufferStatusEXT(GL_FRAMEBUFFER_EXT);
>> + printf("Testing using fb srgb format");
>> + if (status != GL_FRAMEBUFFER_COMPLETE_EXT) {
>> + printf(" - fbo incomplete (status = %s)\n",
>> + piglit_get_gl_enum_name(status));
>> + piglit_report_subtest_result(PIGLIT_SKIP, "srgb fb");
>> + return PIGLIT_SKIP;
>> + }
>> + printf("\n");
>> +
>> + glEnable(GL_TEXTURE_2D);
>> + glBindTexture(GL_TEXTURE_2D, texsrgb);
>> + piglit_draw_rect_tex(-1, -1, 2, 2,
>> + 0, 0, 1, 1);
>> + glDisable(GL_TEXTURE_2D);
>> +
>> + tolerance = 0.6f; /* as required by d3d10 */
>> + /* TODO: should test this tolerance too really right now only test
>> + the previously converted from srgb values, so we only enforce
>> + any value stays the same when doing srgb->linear->srgb, which
>> + corresponds to 0.5 tolerance but only for these exact values. */
>> + glReadPixels(0, 0, 16, 16, GL_RGBA, GL_UNSIGNED_BYTE, &readb[0][0]);
>> +
>> + for (i = 0; i < 256; i++) {
>> +// printf("observed: %d expected: %d\n", readb[i][0], i);
>> + if (readb[i][0] != i) {
>> + printf(" when testing srgb->float->srgb result\n");
>
> printf(" *failed* when testing ....
Ok.
>
>> + pass = GL_FALSE;
>> + break;
>> + }
>> + }
>> +
>> + piglit_present_results();
>> +
>> + piglit_report_subtest_result(pass ? PIGLIT_PASS : PIGLIT_FAIL,
>> + "srgb->linear->srgb");
>> +
>> + glDeleteTextures(1, &texfb);
>> + glDeleteTextures(1, &texsrgb);
>> + glDeleteFramebuffersEXT(1, &fb);
>> +
>> + return pass ? PIGLIT_PASS : PIGLIT_FAIL;
>> +}
>> +
>> +enum piglit_result piglit_display(void)
>> +{
>> + return test_format();
>> +}
>> +
>> +void piglit_init(int argc, char **argv)
>> +{
>> + piglit_require_extension("GL_EXT_framebuffer_object");
>> + piglit_require_extension("GL_ARB_texture_env_combine");
>> + piglit_require_extension("GL_EXT_texture_sRGB");
>> + piglit_require_extension("GL_ARB_framebuffer_sRGB");
>> + piglit_require_extension("GL_ARB_color_buffer_float");
>> + glDisable(GL_DITHER);
>> +}
>> --
>> 1.7.9.5
>
> Otherwise looks good to me.
>
> Don't know if there is a better category for this test.
Yeah I didn't spot anything obvious. Not really texturing nor fbo.
spec/arb_framebuffer_srgb might be an option (though of course this
really is intended to test both the textuing as well as the framebuffer
side conversions). I'll move it there.
Roland
More information about the Piglit
mailing list