It's really a bug.
Because of the HW limitation, vertical stride is at least aligned to 2.
For 1D array image, the data has interval. The size calculated in memset
is right,
but the image size is just twice as big as it.
Use clEnqueueWriteImage is safe and I will fix it later.
On 2015年02月04日 07:11, Rebecca N. Palmer wrote:
Both [3.18 kernel] and the 3.16 kernel have a different intermittent
failure I have
yet to investigate:
compiler_fill_image_1d_array() [FAILED]
Error: dst[j*w + i] == 0
at file
/home/rnpalmer/Debian/builds/stackbuild/beignet/utests/compiler_fill_image_1d_array.cpp,
function compiler_fill_image_1d_array, line 63
The kernel itself (fill j<7,i<32 with 0x03020100) is working
correctly; the problem is that the initial memset() clear (line 30)
sometimes only clears half the array (exactly half, i.e. j<4 is filled
with 0, but the rest is left at whatever it was before).
_______________________________________________
Beignet mailing list
Beignet@lists.freedesktop.org
http://lists.freedesktop.org/mailman/listinfo/beignet
_______________________________________________
Beignet mailing list
Beignet@lists.freedesktop.org
http://lists.freedesktop.org/mailman/listinfo/beignet