[Piglit] [PATCH v2 4/4] getteximage-simple: Stress an i965 blitter path

[Lionel Landwerlin]

On 19/06/18 18:53, Nanley Chery wrote:
> On Tue, Jun 19, 2018 at 11:03:17AM +0100, Lionel Landwerlin wrote:
>> On 18/06/18 19:57, Nanley Chery wrote:
>>> Change the dimension of the texture to test how i965 handles having a
>>> row pitch too large for the BLT engine.
>>>
>>> v2: Query the maximum supported texture width (Ilia Mirkin)
>>> ---
>>>    tests/texturing/getteximage-simple.c | 25 +++++++++++++++++--------
>>>    1 file changed, 17 insertions(+), 8 deletions(-)
>>>
>>> diff --git a/tests/texturing/getteximage-simple.c b/tests/texturing/getteximage-simple.c
>>> index 525e6a392..5b87b7ccd 100644
>>> --- a/tests/texturing/getteximage-simple.c
>>> +++ b/tests/texturing/getteximage-simple.c
>>> @@ -8,6 +8,10 @@
>>>     * texture images is executed before the readback.
>>>     *
>>>     * This used to crash for R300+bufmgr.
>>> + *
>>> + * This also used to stress test the blit methods in i965. The BLT engine only
>>> + * supports pitch sizes up to but not including 32768 dwords. BLORP supports
>>> + * even larger sizes.
>>>     */
>>>    #include "piglit-util-gl.h"
>>> @@ -21,10 +25,10 @@ PIGLIT_GL_TEST_CONFIG_BEGIN
>>>    PIGLIT_GL_TEST_CONFIG_END
>>> -#define MAX_TYPE_VAL UINT8_MAX
>>> -#define PIX_TYPE GLubyte
>>> -#define TEX_TYPE GL_UNSIGNED_BYTE
>>> -#define TEX_INT_FMT GL_RGBA8
>>> +#define MAX_TYPE_VAL 1.0
>>> +#define PIX_TYPE GLfloat
>>> +#define TEX_TYPE GL_FLOAT
>>> +#define TEX_INT_FMT GL_RGBA32F
>>>    #define TEX_FMT GL_RGBA
>>>    #define CHANNELS_PER_PIXEL 4
>> This is changing the format of the texture we're testing with.
> Yes, a higher resolution format is needed to hit the row pitch of
> interest on the older gens which have a max width of 8K.
>
>> In patch 3 I thought you wanted to make this configurable (through arguments
>> to the test maybe?), but this is just switching the format.
> In patch 3, I attempted to make the test configurable via a group of
> #defines.
>
>> Why not add a command line parameter?
>>
> I didn't want to add a command line parameter because there didn't seem
> to be any benefit to doing so. By changing the format and width, we're
> able to continue testing the issue in R300 as well as the teximage
> upload and download paths of interest in i965. Thoughts?

Not really, I was just wondering whether that format change would mean 
that some driver wouldn't get tested anymore.
I guess it's a common enough format that everybody probably supports it.

Thanks for the explanation :


Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin at intel.com>

>
> -Nanley
>
>> Thanks,
>>
>> -
>> Lionel
>>
>>> @@ -42,8 +46,8 @@ static bool test_getteximage(PIX_TYPE *data, size_t data_size, GLint w, GLint h)
>>>    			const unsigned pixel_channel = i % CHANNELS_PER_PIXEL;
>>>    			printf("GetTexImage() returns incorrect data in element %i\n", i);
>>>    			printf("    corresponding to (%i,%i) channel %i\n", pixel % w, pixel / w, pixel_channel);
>>> -			printf("    expected: %i\n", data[i]);
>>> -			printf("    got: %i\n", compare[i]);
>>> +			printf("    expected: %f\n", data[i]);
>>> +			printf("    got: %f\n", compare[i]);
>>>    			match = false;
>>>    			break;
>>>    		}
>>> @@ -53,11 +57,13 @@ static bool test_getteximage(PIX_TYPE *data, size_t data_size, GLint w, GLint h)
>>>    	return match;
>>>    }
>>> +
>>>    enum piglit_result
>>>    piglit_display(void)
>>>    {
>>> -	GLsizei height = 16;
>>> -	GLsizei width = 64;
>>> +	GLsizei height = 2;
>>> +	GLsizei width;
>>> +	glGetIntegerv(GL_MAX_TEXTURE_SIZE, &width);
>>>    	/* Upload random data to a texture with the given dimensions */
>>>    	const unsigned data_channels = width * height * CHANNELS_PER_PIXEL;
>>> @@ -94,6 +100,9 @@ piglit_display(void)
>>>    void piglit_init(int argc, char **argv)
>>>    {
>>> +	if (TEX_TYPE == GL_FLOAT)
>>> +		piglit_require_extension("GL_ARB_texture_float");
>>> +
>>>    	GLuint tex;
>>>    	glGenTextures(1, &tex);
>>