From: Junyan He
Because just platform after BDW will support double,
the special instruction for double MOV is not needed
anymore.
Signed-off-by: Junyan He
---
backend/src/backend/gen75_encoder.cpp | 36 -
From: Junyan He
When we do conversion, the src may be uniform but dst
is not. In this case, we need to set the simd=1.
Signed-off-by: Junyan He
---
backend/src/backend/gen_insn_selection.cpp | 33 ++
1 file
From: Junyan He
Double is supported on BDW later platforms, just normal MOV
can handle the loading of double. So no need for LOAD_DF_IMM
anymore.
Signed-off-by: Junyan He
---
backend/src/backend/gen75_encoder.cpp | 30
From: Junyan He
Signed-off-by: Junyan He
---
utests/compiler_double_convert.cpp | 3 +--
utests/compiler_double_div.cpp | 3 +++
2 files changed, 4 insertions(+), 2 deletions(-)
diff --git a/utests/compiler_double_convert.cpp
From: Junyan He
Signed-off-by: Junyan He
---
backend/src/backend/gen_insn_selection.cpp | 197 +++--
1 file changed, 189 insertions(+), 8 deletions(-)
diff --git a/backend/src/backend/gen_insn_selection.cpp
Would this be better if you turn off the overcommit via proc fs?
Only if you also disable any swap space ( sudo swapoff -a && sudo sh -c
"echo -n 2 > /proc/sys/vm/overcommit_memory" #warning, this may itself
crash your desktop); if I disable overcommit but leave swap on, I get a
hang with
V2:
Fix uniform bug in conversion.
Delete verbose printf in utests.
Fix a bug for BSW when convert half to double.
On Thu, Nov 05, 2015 at 04:15:41PM +0800, junyan...@inbox.com wrote:
> Date: Thu, 5 Nov 2015 16:15:41 +0800
> From: junyan...@inbox.com
> To: beignet@lists.freedesktop.org
>
v2:
1. Just upload the first vme_state.
2. Remove duplicated code in check_opt1_extension.
3. Check image format before cl_gpgpu_bind_image_for_vme.
4. Fix error of getting mv. Because we suppose this kernel run in SIMD16
mode, so dword 0 of grf 1 should be
__gen_ocl_region(8,vme_result.s0),
If the CL device does not support this builtin kernel, the test returns
PASS.
Signed-off-by: Guo Yejun
---
utests/CMakeLists.txt | 1 +
.../builtin_kernel_block_motion_estimate_intel.cpp | 109 +
utests/utest_helper.hpp
__gen_ocl_vme is used for hardware accelerated video motion estimation.
It gets payload values as parameters and uses MOV to pass these payload
values to VME SEND Message's payload grfs. The int8 return value is used
to store SEND Message writeback.
v2:
Remove unnecessary 5 parameters(src_grf*)
v3:
Fix two typos.
Signed-off-by: Chuanbo Weng
---
docs/Beignet.mdwn | 1 +
docs/howto/video-motion-estimation-howto.mdwn | 79 +++
2 files changed, 80 insertions(+)
create mode 100644
Ping for pushed.
-Original Message-
From: Luo, Xionghu
Sent: Wednesday, October 28, 2015 9:42 AM
To: Pan, Xiuli ; beignet@lists.freedesktop.org
Cc: Pan, Xiuli
Subject: RE: [Beignet] [PATCH] utests: fix image_from_buffer bugs
This patch LGTM.
Ping for review.
-Original Message-
From: Beignet [mailto:beignet-boun...@lists.freedesktop.org] On Behalf Of Pan
Xiuli
Sent: Tuesday, November 3, 2015 11:30 AM
To: beignet@lists.freedesktop.org
Cc: Pan, Xiuli
Subject: [Beignet] [PATCH] GBE: fix printf class static
Use wait function to extend a debug function:
void debugwait(void)
This function can hang the gpu unless gpu reset
or host send something to let it go.
EXTREMELY DANGEROUS for machines turn off hangcheck
v2:
Fix some bugs, and add setting predicate and execwidth,
also modify some inst
There are 3 notification can be used by wait, so we
should be able to choose which one we'd like to use.
Also the 3 reg is n0.0 n0.1 and n0.2 so also change
the function name.
Signed-off-by: Pan Xiuli
---
backend/src/backend/gen_encoder.cpp | 4 ++--
Ping for review, thanks!
-Original Message-
From: Beignet [mailto:beignet-boun...@lists.freedesktop.org] On Behalf Of Weng,
Chuanbo
Sent: Tuesday, October 27, 2015 10:16
To: beignet@lists.freedesktop.org
Cc: Wu, Zhiwen
Subject: Re: [Beignet] [PATCH v2 1/2] Add extension
Ping for review.
-Original Message-
From: Pan, Xiuli
Sent: Thursday, October 29, 2015 1:47 PM
To: beignet@lists.freedesktop.org
Cc: Pan, Xiuli
Subject: [PATCH] utests: fix compiler_fill_image_2d_array random bug
Use safer image write instead of map and memset. When
Ping for pushed.
-Original Message-
From: Pan, Xiuli
Sent: Wednesday, November 4, 2015 9:48 AM
To: Rebecca N. Palmer ; beignet@lists.freedesktop.org
Subject: RE: [Beignet] [PATCH v3] GBE: Don't read past end of printf format
string
LGTM, Thanks for your help!
Thread_local is not needed to pass data from one llvm pass to another.
You can still access the info after pass that has already run.
In a later llvm pass, you can use getAnalysis() to get the the
PrintfParser pass handle.
Then expose an interface in PrintfParser like
map
Yes, but the problem is that if two thread has kernel with printf functions,
the map printfs will be cleared in
construction
and destructor. This will cause the one who is still need info in printfs get
null pointer.
thread_local now is to protect printfs from other thread but not pass data from
Hi Ruiling,
As we discussed before, I have refined code to handle both simd8 and
simd16 in backend and ocl kernel. Please
confirm if no problem so that this patchset can be pushed. Thanks!
-Original Message-
From: Weng, Chuanbo
Sent: Friday, November 06, 2015 11:28
To:
The patchset LGTM, pushed, thanks.
> -Original Message-
> From: Beignet [mailto:beignet-boun...@lists.freedesktop.org] On Behalf Of
> He Junyan
> Sent: Thursday, November 5, 2015 16:21
> To: beignet@lists.freedesktop.org
> Subject: Re: [Beignet] [PATCH 1/7 V2] Backend: Delete the useless
Pushed.
> -Original Message-
> From: Pan, Xiuli
> Sent: Friday, November 6, 2015 9:43
> To: Luo, Xionghu; beignet@lists.freedesktop.org
> Cc: Yang, Rong R
> Subject: RE: [Beignet] [PATCH] utests: fix image_from_buffer bugs
>
> Ping for pushed.
>
> -Original Message-
> From: Luo,
Pushed.
> -Original Message-
> From: Beignet [mailto:beignet-boun...@lists.freedesktop.org] On Behalf Of
> Song, Ruiling
> Sent: Wednesday, November 4, 2015 16:42
> To: Meng, Mengmeng; beignet@lists.freedesktop.org
> Cc: Meng, Mengmeng
> Subject: Re: [Beignet] [PATCH v2 1/2] add benckmark
Pushed.
> -Original Message-
> From: Beignet [mailto:beignet-boun...@lists.freedesktop.org] On Behalf Of
> Ruiling Song
> Sent: Wednesday, November 4, 2015 15:50
> To: beignet@lists.freedesktop.org
> Cc: Song, Ruiling
> Subject: [Beignet] [PATCH V3] GBE: Refine ir for memory operation
Thanks. Backported to Fedora. Works perfectly.
On Wed, Nov 4, 2015, 11:48 PM Rebecca N. Palmer
wrote:
> Specifically (and assuming it is the same bug on Ivy Bridge and
> Haswell), fixed in git master by
>
>
LGTM, thanks, pushed.
> -Original Message-
> From: Beignet [mailto:beignet-boun...@lists.freedesktop.org] On Behalf Of
> Pan Xiuli
> Sent: Thursday, October 29, 2015 13:47
> To: beignet@lists.freedesktop.org
> Cc: Pan, Xiuli
> Subject: [Beignet] [PATCH] utests: fix
Pushed.
> -Original Message-
> From: Beignet [mailto:beignet-boun...@lists.freedesktop.org] On Behalf Of
> Pan, Xiuli
> Sent: Friday, November 6, 2015 9:44
> To: 'Rebecca N. Palmer'; 'beignet@lists.freedesktop.org'
> Subject: Re: [Beignet] [PATCH v3] GBE: Don't read past end of printf
28 matches
Mail list logo