[Piglit] [PATCH] util: specify channel sizes for choose_config_attribs()
Brian Paul
brianp at vmware.com
Tue Jun 21 17:20:17 UTC 2016
On 06/20/2016 10:25 AM, Jose Fonseca wrote:
> On 17/06/16 21:46, Brian Paul wrote:
>> WAFFLE_RED_SIZE=1 (for example) gets passed to glXChooseFBConfig as
>> GLX_RED_SIZE=1. GLX chooses the deepest pixel format which meets that
>> minimum value so we typically get an 8 bit/channel format.
>>
>> But with wglChoosePixelFormatARB() we may get the shallowest pixel
>> format that meets that value so we may wind up getting a 5/6/5 format.
>> Many piglit tests aren't prepared for that and report failures because
>> the color tolerances are too tight. In any case, we really want to test
>> 8-bit formats.
>>
>> This change tells waffle to first try 8/8/8 color, 24-bit Z. And if
>> that fails, try 5/6/5 with 16-bit Z.
>> ---
>> .../piglit-framework-gl/piglit_winsys_framework.c | 28
>> ++++++++++++++++------
>> 1 file changed, 21 insertions(+), 7 deletions(-)
>>
>> diff --git a/tests/util/piglit-framework-gl/piglit_winsys_framework.c
>> b/tests/util/piglit-framework-gl/piglit_winsys_framework.c
>> index ff3428a..315d6de 100644
>> --- a/tests/util/piglit-framework-gl/piglit_winsys_framework.c
>> +++ b/tests/util/piglit-framework-gl/piglit_winsys_framework.c
>> @@ -102,8 +102,15 @@ post_redisplay(struct piglit_gl_framework *gl_fw)
>> piglit_winsys_framework(gl_fw)->need_redisplay = true;
>> }
>>
>> +/**
>> + * Create a integer array of WAFFLE_x attribute/value pairs from
>> + * the given test_config (which specifies color, depth, samples, etc.
>> + * \param min_color minimum bits per channel for color components.
>> + * \param min_z minimum bits for depth/Z buffer.
>> + */
>> static const int32_t*
>> -choose_config_attribs(const struct piglit_gl_test_config *test_config)
>> +choose_config_attribs(const struct piglit_gl_test_config *test_config,
>> + int min_color, int min_z)
>> {
>> static int32_t attrib_list[64];
>> int i = 0;
>> @@ -112,21 +119,21 @@ choose_config_attribs(const struct
>> piglit_gl_test_config *test_config)
>> if (test_config->window_visual &
>> (PIGLIT_GL_VISUAL_RGB | PIGLIT_GL_VISUAL_RGBA)) {
>> attrib_list[i++] = WAFFLE_RED_SIZE;
>> - attrib_list[i++] = 1;
>> + attrib_list[i++] = min_color;
>> attrib_list[i++] = WAFFLE_GREEN_SIZE;
>> - attrib_list[i++] = 1;
>> + attrib_list[i++] = min_color;
>> attrib_list[i++] = WAFFLE_BLUE_SIZE;
>> - attrib_list[i++] = 1;
>> + attrib_list[i++] = min_color;
>> }
>>
>> if (test_config->window_visual & PIGLIT_GL_VISUAL_RGBA) {
>> attrib_list[i++] = WAFFLE_ALPHA_SIZE;
>> - attrib_list[i++] = 1;
>> + attrib_list[i++] = min_color;
>> }
>>
>> if (test_config->window_visual & PIGLIT_GL_VISUAL_DEPTH) {
>> attrib_list[i++] = WAFFLE_DEPTH_SIZE;
>> - attrib_list[i++] = 1;
>> + attrib_list[i++] = min_z;
>> }
>>
>> if (test_config->window_visual & PIGLIT_GL_VISUAL_STENCIL) {
>> @@ -198,8 +205,15 @@ piglit_winsys_framework_init(struct
>> piglit_winsys_framework *winsys_fw,
>> struct piglit_gl_framework *gl_fw = &wfl_fw->gl_fw;
>> bool ok = true;
>>
>> + // First try to get 8-bit color channels, 24-bit Z
>> ok = piglit_wfl_framework_init(wfl_fw, test_config, platform,
>> - choose_config_attribs(test_config));
>> + choose_config_attribs(test_config, 8, 24));
>> + if (!ok) {
>> + // Try shallower color/Z. We may get 565 color and 16-bit Z.
>> + ok = piglit_wfl_framework_init(wfl_fw, test_config, platform,
>> + choose_config_attribs(test_config, 1, 1));
>> + }
>> +
>> if (!ok)
>> goto fail;
>>
>>
>
>
> Seems a good idea to me.
>
> Reviewed-by: Jose Fonseca <jfonseca at vmware.com>
>
>
> Given certain tests appear to rely on having 8 bit precision, I think we
> should warn, and possibly even fail, when we don't get such visuals.
> Anything but silently fallback to lower precision. We could maybe even
> check things like GL_RED_BITS / GL_GREEN_BITS / GL_BLUE_BITS for extra
> assurance, which could be done in a place agnostic to Waffle/GLUT/etc.
Yeah, we should probably at least print a warning when we don't get
8-bit color.
If we do get a 5/6/5 format, the other thing we could do is set the
piglit_tolerance[] values larger by calling piglit_set_tolerance_for_bits().
I'll take a closer look, but commit this patch in the mean time.
-Brian
More information about the Piglit
mailing list