Re: [Beignet] [Patch V2] Refine the intel gpgpu delete.

2014-10-27 Thread Zhigang Gong
LGTM, just pushed, thanks. On Mon, Oct 27, 2014 at 04:16:51PM +0800, Yang Rong wrote: > The intel gpgpu struct is destroyed when a new gpgpu struct needed. But in > that time, > the command batch relative with the destroyed gpgpu may have not finish, and > the resource > in gpgpu still used by g

Re: [Beignet] [Intel-gfx] Beignet crashes on vanilla 3.17.1 with IVB hardware

2014-10-27 Thread Zhigang Gong
Vasily, Could you try ImageMagick(convert) again with latest git master beignet(git-e46764f). It should work now. Thanks, Zhigang Gong. On Fri, Oct 24, 2014 at 04:36:49PM +0300, Vasily Khoruzhick wrote: > Hi Zhigang, > > On Fri, Oct 24, 2014 at 12:13 PM, Zhigang Gong > wrote: > > Hi, > > > > L

Re: [Beignet] [PATCH] Fix a size assert when setup bti.

2014-10-27 Thread Zhigang Gong
Is there anybody can verify this patch on BDW? Thanks. On Mon, Oct 20, 2014 at 03:46:17PM +0800, Yang Rong wrote: > Global constant buffer size is not align to 4 byte, will cause assert in BDW > when set bti. > Per spec, the low two bits of surface state's width must be 11 if > SURFACE_BUFFER's

Re: [Beignet] [PATCH] Fix the compare_image_2d_and_1d_array test case bug

2014-10-27 Thread Zhigang Gong
LGTM, will push latter, thanks. On Mon, Oct 27, 2014 at 03:34:11PM +0800, junyan...@inbox.com wrote: > From: Junyan He > > The test case use OCL_MAP_BUFFER_GTT to map the image > buffers and then do the result comparison, which may > cause problems. > On IVB and HSW, the slice pitch is equal but

Re: [Beignet] [patch v2 1/2] add llvm Intrinsic call support.

2014-10-27 Thread Zhigang Gong
This version LGTM, will push latter, thanks. On Mon, Oct 27, 2014 at 11:14:49AM +0800, xionghu@intel.com wrote: > From: Luo Xionghu > > so far this patch only support uadd_with_overflow. > sadd_with_overflow, usub_with_overflow, smul_with_overflow, > umul_with_overflow, bswap are not suppor

Re: [Beignet] [PATCH 3/5] Remove intel_gpgpu_check_binded_buf_address()

2014-10-27 Thread Zhenyu Wang
On 2014.10.27 12:57:59 +0800, Zhenyu Wang wrote: > On 2014.10.24 18:10:36 +0800, Zhigang Gong wrote: > > This assertion is just to make sure we will not get a NULL pointer > > for a normal buffer. The OpenCL spec doesn't give a very specific > > statement about a NULL buffer object. But it does all

[Beignet] [patch v2 1/2] add llvm Intrinsic call support.

2014-10-27 Thread xionghu . luo
From: Luo Xionghu so far this patch only support uadd_with_overflow. sadd_with_overflow, usub_with_overflow, smul_with_overflow, umul_with_overflow, bswap are not supported yet. this funtion should be implemented by carrier flag later. v2: update comments. Signed-off-by: Luo Xionghu --- back

[Beignet] [patch v2 2/2] add utest compiler_overflow for llvm intrinsic function.

2014-10-27 Thread xionghu . luo
From: Luo Xionghu this case only runs for uadd_with_over_flow function so far. Signed-off-by: Luo Xionghu --- kernels/compiler_overflow.cl | 18 ++ utests/CMakeLists.txt|1 + utests/compiler_overflow.cpp | 56 ++ 3 files chang

Re: [Beignet] [PATCH] Refine the intel gpgpu delete.

2014-10-27 Thread Yang, Rong R
Yes, I assume that all the gpgpu in the list would complete in order, but it may not true. I have sent a new version. > -Original Message- > From: Zhigang Gong [mailto:zhigang.g...@linux.intel.com] > Sent: Monday, October 27, 2014 14:33 > To: Yang, Rong R > Cc: beignet@lists.freedesktop.o

[Beignet] [Patch V2] Refine the intel gpgpu delete.

2014-10-27 Thread Yang Rong
The intel gpgpu struct is destroyed when a new gpgpu struct needed. But in that time, the command batch relative with the destroyed gpgpu may have not finish, and the resource in gpgpu still used by gpgpu, can't be destroyed. So, when delete a gpgpu, check the batch status, if have not complete,

[Beignet] [PATCH] Fix the compare_image_2d_and_1d_array test case bug

2014-10-27 Thread junyan . he
From: Junyan He The test case use OCL_MAP_BUFFER_GTT to map the image buffers and then do the result comparison, which may cause problems. On IVB and HSW, the slice pitch is equal but on BDW, because we change the slice pitch of image array, it cause this bug. Modify it by using the standard clEn

Re: [Beignet] [PATCH] Refine the intel gpgpu delete.

2014-10-27 Thread Zhigang Gong
When loop for the gpgpu list, it's better to check each gpgpu structure once with non-blocking mode. Please see below diff. diff --git a/src/intel/intel_gpgpu.c b/src/intel/intel_gpgpu.c index 7f9b8ac..c4a1c9c 100644 --- a/src/intel/intel_gpgpu.c +++ b/src/intel/intel_gpgpu.c @@ -168,14 +168,17 @

Re: [Beignet] possilbe bug when run opencv_test_imgproc

2014-10-27 Thread yan . wang
I tried it and the issue is still exist. Yan Wang > Hi, Yan, > > I have sent a patch "Refine the intel gpgpu delete. ", it may fix some > race condition fail, can you try it? > >> -Original Message- >> From: Beignet [mailto:beignet-boun...@lists.freedesktop.org] On Behalf >> Of >> yan