[Beignet] OpenCV Morph OCL Perf case is crash based on beignet

/yanwang/beignet/backend/src/llvm/llvm_legalize.cpp, function bool gbe::Legalize::legalizeFunction(llvm::Function), line 516 After asked Junyan, it seems phi op code hasn't been supported. Thanks. Yan Wang ___ Beignet mailing list Beignet

Re: [Beignet] OpenCV Morph OCL Perf case is crash based on beignet

Sorry, I should update beignet not opencv. I will try. Yan Wang You commit seems not to be merged yet. If merged, I will try. Thanks. Yan Wang Did you try latest git master? It should has been fixed by commit: commit 5417a7e2afb3f2043dbb8abbaebca5115b87ab70 Author: Zhigang Gong

[Beignet] opencv_test_imgproc crashs when test OCL_Imgproc/HoughLines.RealImage/0

(const gbe::ir::BasicBlock*, gbe::ir::Register), line 91 Thanks. Yan Wang ___ Beignet mailing list Beignet@lists.freedesktop.org http://lists.freedesktop.org/mailman/listinfo/beignet

Re: [Beignet] opencv_test_imgproc crashs when test OCL_Imgproc/HoughLines.RealImage/0

, function gbe::ir::LiveOutSet::RegDefSet gbe::ir::LiveOutSet::getDefSet(const gbe::ir::BasicBlock*, gbe::ir::Register), line 91 Thanks. Yan Wang ___ Beignet mailing list Beignet@lists.freedesktop.org http://lists.freedesktop.org/mailman/listinfo/beignet

Re: [Beignet] opencv_test_imgproc crashs when test OCL_Imgproc/HoughLines.RealImage/0

!= bbIt-second-end() defIt-second != NULL at file /home/yanwang/beignet/backend/src/ir/value.cpp, function gbe::ir::LiveOutSet::RegDefSet gbe::ir::LiveOutSet::getDefSet(const gbe::ir::BasicBlock*, gbe::ir::Register), line 91 Thanks. Yan Wang

Re: [Beignet] OpenCV Morph OCL Perf case is crash based on beignet

I confirmed it is fixed. Thanks. Yan, How about this one? On Thu, Oct 23, 2014 at 12:29:34AM -0700, yan.w...@linux.intel.com wrote: Sorry, I should update beignet not opencv. I will try. Yan Wang You commit seems not to be merged yet. If merged, I will try. Thanks. Yan Wang

Re: [Beignet] opencv_test_imgproc crashs when test OCL_Imgproc/HoughLines.RealImage/0

/yanwang/beignet/backend/src/ir/value.cpp, function gbe::ir::LiveOutSet::RegDefSet gbe::ir::LiveOutSet::getDefSet(const gbe::ir::BasicBlock*, gbe::ir::Register), line 91 Thanks. Yan Wang ___ Beignet mailing list Beignet

Re: [Beignet] opencv_test_imgproc crashs when test OCL_Imgproc/HoughLines.RealImage/0

/yanwang/beignet/backend/src/ir/value.cpp, function gbe::ir::LiveOutSet::RegDefSet gbe::ir::LiveOutSet::getDefSet(const gbe::ir::BasicBlock*, gbe::ir::Register), line 91 Thanks. Yan Wang ___ Beignet mailing list Beignet

Re: [Beignet] opencv_test_imgproc crashs when test OCL_Imgproc/HoughLines.RealImage/0

at file /home/yanwang/beignet/backend/src/ir/value.cpp, function gbe::ir::LiveOutSet::RegDefSet gbe::ir::LiveOutSet::getDefSet(const gbe::ir::BasicBlock*, gbe::ir::Register), line 91 Thanks. Yan Wang ___ Beignet mailing list

[Beignet] possilbe bug when run opencv_test_imgproc

influence it. Thanks. Yan Wang ___ Beignet mailing list Beignet@lists.freedesktop.org http://lists.freedesktop.org/mailman/listinfo/beignet

Re: [Beignet] possilbe bug when run opencv_test_imgproc

In my baytail T platform, it is still exist. Before leaving office, I have built the latest beignet. I will try it again tomorrow. Thanks. Yan Wang Can't reproduce it here. Did you use the latest git master? The latest git version is commit 33a043e903f9d61445d3fb7972becf83c2842276

Re: [Beignet] possilbe bug when run opencv_test_imgproc

Update: Under root account, it will crash. Under user account(e.g. yanwang), it will pass. In my baytail T platform, it is still exist. Before leaving office, I have built the latest beignet. I will try it again tomorrow. Thanks. Yan Wang Can't reproduce it here. Did you use the latest

Re: [Beignet] possilbe bug when run opencv_test_imgproc

implementation in OpenCV. Thanks. Yan Wang Hi, All, I found one possible bug for review. if run the following: ./opencv_test_imgproc --gtest_filter=OCL_ImageProc/Filter2D.Mat*. OCL_ImageProc/Filter2D.Mat/256 failed and continue. But the whole test flow will crash in OCL_ImageProc

Re: [Beignet] possilbe bug when run opencv_test_imgproc

] OCL_ImgProc/Canny.Accuracy/10, where GetParam() = (Channels(3), AppertureSize(3), L2gradient(true), UseRoi(false)) [ FAILED ] OCL_Imgproc/HoughLines.RealImage/2, where GetParam() = (1, 0.00872665, 80) Yan Wang I have BYT box, an IVB machine and a HSW notebook. All of them haven't this issue

Re: [Beignet] possilbe bug when run opencv_test_imgproc

Could you give me one URL? Thanks. Yan Wang All of these three failures are already tracked in JIRA. If you have access to JIRA, you can check them easily. Thanks, Zhigang Gong. On Fri, Oct 24, 2014 at 10:33 PM, yan.w...@linux.intel.com wrote: Sure. I could try Yang Rong's patch. BTW

[Beignet] [PATCH] Implement cl_khr_image2d_from_buffer extension.

--- CMakeLists.txt | 2 + kernels/image_2D_buffer.cl | 15 + src/CMakeLists.txt | 5 ++ src/cl_api.c | 9 +++ src/cl_device_id.c | 4 ++ src/cl_device_id.h | 4 ++ src/cl_extensions.c| 2 +- src/cl_gt_device.h |

[Beignet] [PATCH] Implement cl_khr_image2d_from_buffer extension.

Implement cl_khr_image2d_from_buffer extension. --- CMakeLists.txt | 2 + kernels/image_2D_buffer.cl | 15 + src/CMakeLists.txt | 5 ++ src/cl_api.c | 9 +++ src/cl_device_id.c | 4 ++ src/cl_device_id.h | 4 ++ src/cl_extensions.c

[Beignet] [PATCH] Fix based on piglit OpenCL falied case (cl-api-compile-program).

1. Return the expected error code. 2. Don't destroy cl_program object after comile error because it may be used still in the future. Signed-off-by: Yan Wang yan.w...@linux.intel.com --- src/cl_program.c | 6 ++ 1 file changed, 2 insertions(+), 4 deletions(-) diff --git a/src/cl_program.c b

[Beignet] [PATCH] Fix based on piglit OpenCL failed case (cl-program-tester).

Fix tests/cl/program/build/optimization-options-cl10.cl After calling check_cl_version_option, -cl-std=CLX.X should be removed. This options couldn't be accepted by the subsequent process. Signed-off-by: Yan Wang yan.w...@linux.intel.com --- src/cl_program.c | 50

Re: [Beignet] [PATCH] Implement cl_khr_image2d_from_buffer extension.

Thanks for your review. Agree your points. Yan Wang Thanks for the patch. But This implementation may hurt performance. The major reason is that the image has some special layout requirement and if we want to create an image from eaxct the pitch * height buffer object, we have to introduce

[Beignet] [PATCH] Fix PrintfState copying.

PrintfState includes std::string object and shouldn't be copied by malloc/memcpy. Signed-off-by: Yan Wang yan.w...@linux.intel.com --- backend/src/ir/printf.hpp | 23 +++ 1 file changed, 19 insertions(+), 4 deletions(-) diff --git a/backend/src/ir/printf.hpp b/backend/src/ir

[Beignet] [PATCH] Fix delete operator using.

The 2 delete operators work on array pointer. Signed-off-by: Yan Wang yan.w...@linux.intel.com --- utests/compiler_fill_gl_image.cpp | 2 +- utests/utest_helper.cpp | 2 +- 2 files changed, 2 insertions(+), 2 deletions(-) diff --git a/utests/compiler_fill_gl_image.cpp b/utests

[Beignet] [PATCH v2] Fix PrintfState copying.

PrintfState include std::string object and shouldn't be copied by malloc/memcpy. Signed-off-by: Yan Wang yan.w...@linux.intel.com --- backend/src/ir/printf.hpp | 33 + 1 file changed, 29 insertions(+), 4 deletions(-) diff --git a/backend/src/ir/printf.hpp b

[Beignet] [PATCH] Fix loop condition of PrintfSet constructor.

Signed-off-by: Yan Wang yan.w...@linux.intel.com --- backend/src/ir/printf.hpp | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/backend/src/ir/printf.hpp b/backend/src/ir/printf.hpp index 3b176f6..cc1f8dc 100644 --- a/backend/src/ir/printf.hpp +++ b/backend/src/ir/printf.hpp

[Beignet] [PATCH] Use CreateCall instead of CreateCall2 because llvm3.7 has removed CreateCall2.

Signed-off-by: Yan Wang <yan.w...@linux.intel.com> --- backend/src/llvm/llvm_profiling.cpp | 18 ++ 1 file changed, 18 insertions(+) diff --git a/backend/src/llvm/llvm_profiling.cpp b/backend/src/llvm/llvm_profiling.cpp index 8c9157c..e548305 100644 --- a/backend/sr

Re: [Beignet] [PATCH] Use CreateCall instead of CreateCall2 because llvm3.7 has removed CreateCall2.

> > >> -Original Message- >> From: Beignet [mailto:beignet-boun...@lists.freedesktop.org] On Behalf >> Of >> Yan Wang >> Sent: Wednesday, November 18, 2015 6:47 PM >> To: beignet@lists.freedesktop.org >> Cc: Yan Wang <yan.w...@linux.inte

[Beignet] [PATCH v2] Use CreateCall instead of CreateCall2.

Signed-off-by: Yan Wang <yan.w...@linux.intel.com> --- backend/src/llvm/llvm_profiling.cpp | 8 1 file changed, 4 insertions(+), 4 deletions(-) diff --git a/backend/src/llvm/llvm_profiling.cpp b/backend/src/llvm/llvm_profiling.cpp index 8c9157c..3fbd00d 100644 --- a/backend/sr

Re: [Beignet] [PATCH v2] Use CreateCall instead of CreateCall2.

So should we rollback to v1? It should be safe because it is only for LLVM >=3.7. Thanks. Yan Wang > Build fail in LLVM3.5.2. > >> -Original Message- >> From: Beignet [mailto:beignet-boun...@lists.freedesktop.org] On Behalf >> Of >> Yan Wang >>

Re: [Beignet] [PATCH v2] Use CreateCall instead of CreateCall2.

Thanks. Yan Wang > The llvm function prototype is CreateCall((Value *Callee, ArrayRef *> Args = None, const Twine = "") > Cast from std::initializer_list to ArrayRef<> is not supported on older > llvm version. > Please try: >/* Add

[Beignet] [PATCH] Add cl_khr_3d_image_writes into info string.

From: Yan Wang <yan.w...@linux.intel.com> The extension is supported in fact and avoid misunderstanding. --- src/cl_extensions.c | 2 ++ 1 file changed, 2 insertions(+) diff --git a/src/cl_extensions.c b/src/cl_extensions.c index 349f2f1..183aafc 100644 --- a/src/cl_extensions.c +++

[Beignet] [Printf v2][PATCH 07/12] Implement emision of printf instruction.

From: Yan Wang <yan.w...@linux.intel.com> Contributor: Junyan He <junyan...@linux.intel.com> Signed-off-by: Yan Wang <yan.w...@linux.intel.com> --- backend/src/llvm/llvm_gen_backend.cpp | 95 +-- 1 file changed, 80 insertions(+), 15 deleti

Re: [Beignet] [Printf v2][PATCH 07/12] Add the implementation of printf ir instruction.

Sorry. I have re-sent 7/12. Yan Wang > patch of 06 and 07 have the same title? > I think it is a typo here. > Please correct it. > All the other things are OK, just rename this one and > the whole patchset can be pushed later. > > Also can push my patch about print

[Beignet] [Printf v2][PATCH 03/12] Reconstruct printf parser.

From: Yan Wang <yan.w...@linux.intel.com> Contributor: Junyan He <junyan...@linux.intel.com> Signed-off-by: Yan Wang <yan.w...@linux.intel.com> --- backend/src/ir/unit.cpp | 1 - backend/src/ir/unit.hpp | 2 +- backend/src/llvm/llvm_gen_b

[Beignet] [Printf v2][PATCH 04/12] Add LLVM fcuntion definition of printf.

From: Yan Wang <yan.w...@linux.intel.com> Contributor: Junyan He <junyan...@linux.intel.com> Signed-off-by: Yan Wang <yan.w...@linux.intel.com> --- backend/src/llvm/llvm_gen_ocl_function.hxx | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-) diff --git

[Beignet] [Printf v2][PATCH 11/12] Output printf result.

From: Yan Wang <yan.w...@linux.intel.com> Contributor: Junyan He <junyan...@linux.intel.com> Signed-off-by: Yan Wang <yan.w...@linux.intel.com> --- backend/src/ir/printf.cpp | 122 +- backend/src/ir/printf.hpp | 2 +- 2 files chang

[Beignet] [Printf v2][PATCH 01/12] Change printf data structure and remove old code.

From: Yan Wang <yan.w...@linux.intel.com> Contributor: Junyan He <junyan...@linux.intel.com> Signed-off-by: Yan Wang <yan.w...@linux.intel.com> --- backend/src/backend/program.cpp | 14 -- backend/src/backend/program.hpp | 10 +- backend/src/gbe_bin_interpr

[Beignet] [Printf v2][PATCH 02/12] Add PrintfLog structure.

From: Yan Wang <yan.w...@linux.intel.com> Contributor: Junyan He <junyan...@linux.intel.com> Signed-off-by: Yan Wang <yan.w...@linux.intel.com> --- backend/src/ir/printf.hpp | 25 + 1 file changed, 25 insertions(+) diff --git a/backend/src/ir/printf.hp

[Beignet] [Printf v2][PATCH 05/12] Add tuple processing logic for printf.

From: Yan Wang <yan.w...@linux.intel.com> Contributor: Junyan He <junyan...@linux.intel.com> Signed-off-by: Yan Wang <yan.w...@linux.intel.com> --- backend/src/ir/context.hpp | 5 + backend/src/ir/function.hpp | 8 2 files changed, 13 insertions(+) diff --gi

[Beignet] [Printf v2][PATCH 09/12] Implement ASM generation of printf.

From: Yan Wang <yan.w...@linux.intel.com> Contributor: Junyan He <junyan...@linux.intel.com> Signed-off-by: Yan Wang <yan.w...@linux.intel.com> --- backend/src/backend/gen8_context.cpp | 36 +++ backend/src/backend/gen8_context.hpp | 1 + backend/src/backe

[Beignet] [Printf v2][PATCH 10/12] Implement printf buffer management.

From: Yan Wang <yan.w...@linux.intel.com> Contributor: Junyan He <junyan...@linux.intel.com> Signed-off-by: Yan Wang <yan.w...@linux.intel.com> --- backend/src/backend/program.cpp | 10 + backend/src/backend/program.h | 12 +- backend/src/backend/program.hpp | 7 --

[Beignet] [Printf v2][PATCH 06/12] Add the implementation of printf ir instruction.

From: Yan Wang <yan.w...@linux.intel.com> Contributor: Junyan He <junyan...@linux.intel.com> Signed-off-by: Yan Wang <yan.w...@linux.intel.com> --- backend/src/ir/instruction.cpp | 57 +- backend/src/ir/instruction.hpp | 13

[Beignet] [Printf v2][PATCH 08/12] Implement instruction selection of printf.

From: Yan Wang <yan.w...@linux.intel.com> Contributor: Junyan He <junyan...@linux.intel.com> Signed-off-by: Yan Wang <yan.w...@linux.intel.com> --- backend/src/backend/gen_context.cpp| 3 + backend/src/backend/gen_context.hpp| 1

[Beignet] [Printf v2][PATCH 07/12] Add the implementation of printf ir instruction.

From: Yan Wang <yan.w...@linux.intel.com> Contributor: Junyan He <junyan...@linux.intel.com> Signed-off-by: Yan Wang <yan.w...@linux.intel.com> --- backend/src/llvm/llvm_gen_backend.cpp | 95 +-- 1 file changed, 80 insertions(+), 15 deleti

Re: [Beignet] [Printf][PATCH 06/11] Implement emision of printf instruction.

Now the root cause has been founded. The allocated surface size is not enough because it is dependent on global size. I Will fix it and resend patch set based on all previous review comments. Thanks. Yan Wang > After applied the printf patch set, I find the last test still > failed, pleas

Re: [Beignet] [PATCH] Fix type assert error generated by lstPartSum incorrect type

After applying this patch, benchmark of workgroup add optimization could run on my BSW platform. Thanks. Yan Wang > Signed-off-by: Grigore Lupescu <grigore.lupe...@intel.com> > --- > backend/src/backend/gen_insn_selection.cpp | 8 > 1 file changed, 4 insertio

[Beignet] [Printf][PATCH 08/11] Implement ASM generation of printf.

Contributor: Junyan He <junyan...@linux.intel.com> Signed-off-by: Yan Wang <yan.w...@linux.intel.com> --- backend/src/backend/gen8_context.cpp | 36 +++ backend/src/backend/gen8_context.hpp | 1 + backend/src/backend/gen_conte

[Beignet] [Printf][PATCH 06/11] Implement emision of printf instruction.

Contributor: Junyan He <junyan...@linux.intel.com> Signed-off-by: Yan Wang <yan.w...@linux.intel.com> --- backend/src/ir/context.hpp| 5 ++ backend/src/llvm/llvm_gen_backend.cpp | 89 --- 2 files changed, 78 insertions(+), 16 deletions(-)

[Beignet] [Printf][PATCH 10/11] Output printf result.

Contributor: Junyan He <junyan...@linux.intel.com> Signed-off-by: Yan Wang <yan.w...@linux.intel.com> --- backend/src/ir/printf.cpp | 122 +- backend/src/ir/printf.hpp | 2 +- 2 files changed, 112 insertions(+), 12 deletions(-) diff --g

[Beignet] [Printf][PATCH 04/11] Add the implementation of printf ir instruction.

Contributor: Junyan He <junyan...@linux.intel.com> Signed-off-by: Yan Wang <yan.w...@linux.intel.com> --- backend/src/ir/function.hpp| 8 ++ backend/src/ir/instruction.cpp | 57 +- backend/src/ir/instruction.hpp | 13 ++ ba

[Beignet] [Printf][PATCH 01/11] Change printf data structure and remove old code.

Contributor: Junyan He <junyan...@linux.intel.com> Signed-off-by: Yan Wang <yan.w...@linux.intel.com> --- backend/src/backend/program.cpp | 14 -- backend/src/backend/program.hpp | 10 +- backend/src/gbe_bin_interpreter.cpp | 2 - backend/src/i

Re: [Beignet] [PATCH] utest: do not check MV near image border

Now this case could passed when previous test_printf case has multiply tests. VME engine seems to read data out of specified image buffer which is based on drm bo. If this drm bo of src/ref image object reuse from previous bo with garbage by coincidence, it will cause different MV results. Yan

[Beignet] [PATCH] Add condition checking of residuals because it may be NULL.

From: Yan Wang <yan.w...@linux.intel.com> --- src/kernels/cl_internal_block_motion_estimate_intel.cl | 9 ++--- 1 file changed, 6 insertions(+), 3 deletions(-) diff --git a/src/kernels/cl_internal_block_motion_estimate_intel.cl b/src/kernels/cl_internal_block_motion_estimate_intel.cl

[Beignet] [PATCH] Remove unncessary assertion in printf processing.

From: Yan Wang <yan.w...@linux.intel.com> It causes alert when printf long vector. --- backend/src/llvm/llvm_gen_backend.cpp | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/backend/src/llvm/llvm_gen_backend.cpp b/backend/src/llvm/llvm_gen_backend.cpp index 51a1dab..7

[Beignet] [PATCH] MAD compact instrcution could not support "absolute" attribute.

From: Yan Wang <yan.w...@linux.intel.com> If absolute of SRCs of MAD instruction is 1, doens't use compact instruction. Signed-off-by: Yan Wang <yan.w...@linux.intel.com> --- backend/src/backend/gen_insn_compact.cpp | 2 ++ 1 file changed, 2 insertions(+) diff --git a/backend

[Beignet] [PATCH] Add read_imagef benchmark for optimization.

From: Yan Wang <yan.w...@linux.intel.com> --- benchmark/CMakeLists.txt | 1 + benchmark/benchmark_read_image_float.cpp | 65 kernels/compiler_read_image_float.cl | 9 + 3 files changed, 75 insertions(+) create mode 100644 ben

Re: [Beignet] [PATCH] Add read_imagef benchmark for optimization.

On Mon, 2016-09-12 at 06:53 +, Yang, Rong R wrote: > > > -Original Message- > > From: Beignet [mailto:beignet-boun...@lists.freedesktop.org] On > > Behalf Of > > yan.w...@linux.intel.com > > Sent: Monday, September 5, 2016 14:52 > > To: beignet

[Beignet] [PATCH] Reduce the compilation time of inline pass in runModulePass().

From: Yan Wang <yan.w...@linux.intel.com> It could reduce much compilation time when run Luxmark scenes. Avoid calling inline pass many times in runModulePass when module is changed by the other pass. Create a single funtion to run inline pass. In this single funtion, lower pass and stric

[Beignet] [PATCH] Fix bug: Initialize bti of LoadInstuctionPattern::shootByteGatherMsg().

From: Yan Wang <yan.w...@linux.intel.com> If it isn't initialized, Luxmark hotel scene will display wrong. --- backend/src/backend/gen_insn_selection.cpp | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-) diff --git a/backend/src/backend/gen_insn_selection.cpp b/backend/src/b

[Beignet] [PATCH] Fix bug: Initialize bti LoadInstuctionPattern::shootUntypedReadMsg().

From: Yan Wang <yan.w...@linux.intel.com> If it isn't initialized, Luxmark hotel scene will display wrong. --- backend/src/backend/gen_insn_selection.cpp | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-) diff --git a/backend/src/backend/gen_insn_selection.cpp b/backend/src/b

[Beignet] [PATCH v2] Fix bug: Initialize bti of LoadInstuctionPattern::shootByteGatherMsg().

From: Yan Wang <yan.w...@linux.intel.com> If it isn't initialized, Luxmark hotel scene will display wrong. --- backend/src/backend/gen_insn_selection.cpp | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/backend/src/backend/gen_insn_selection.cpp b/backend/src/b

[Beignet] [PATCH] Fix getting bitwidth of PointerType of LLVM.

From: Yan Wang <yan.w...@linux.intel.com> PointerType could not be forced to IntegerTyoe for getting bitwidth. With Rong's comments, use getTypeBitSize() instead of Type::getIntegerBitWidth(). --- backend/src/llvm/llvm_gen_backend.cpp | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)

[Beignet] [PATCH] Restore jump threading pass for reducing compiling time when run the large and complex kernel like Luxmark.

From: Yan Wang <yan.w...@linux.intel.com> Jump threading pass could optimize the connection between LLVM basic blocks of the function and provide the chance to merge and remove unnecessary basic blocks to reduce the compilation time and ASM code size. Signed-off-by: Yan Wang

Re: [Beignet] [PATCH 1/2] remove some redundant code for printf

LGTM. Thanks. Yan Wang On Mon, 2016-11-21 at 18:16 +0800, Guo, Yejun wrote: > tmp0 is added into src in selection stage, and just ignored at > context > stage, it is redundant. > > Signed-off-by: Guo, Yejun <yejun@intel.com> > --- > backend/src/backend/

Re: [Beignet] [PATCH 2/2] do not care dst for printf

LGTM. Thanks. Yan Wang On Mon, 2016-11-21 at 18:16 +0800, Guo, Yejun wrote: > acutally, the dst of printf means nothing, don't need to touch it. > > Signed-off-by: Guo, Yejun <yejun@intel.com> > --- > backend/src/backend/gen_context.cpp| 14 ++ &

[Beignet] [PATCH] Avoid possible invalid pointer by vector interator.

From: Yan Wang <yan.w...@linux.intel.com> "revisit" as vector containber will be pushed more elements in findPointerEsacape() and cause previous interator to introduce possible invalid pointer. When compiling huge kernel like blender, it will cause random segment fault cr

[Beignet] [PATCH] Avoid possible invalid pointer by vector interator.

From: Yan Wang <yan.w...@linux.intel.com> "revisit" as vector containber will be pushed more elements in findPointerEsacape() and cause previous interator to introduce possible invalid pointer. When compiling huge kernel like blender, it will cause random segment fault cr

Re: [Beignet] [PATCH] GBE: reorder the LLVM pass to reduce the compilation time.

LGTM. Thanks. Yan Wang On Fri, 2016-12-16 at 16:38 +0800, Yang Rong wrote: > Set all function's linkage to LinkOnceAnyLinkage, then Inlining pass > could delete the inlined functions. > And reorder createFunctionInliningPass before > createStripAttributesPass > can reduce the c

Re: [Beignet] [PATCH v2] Provide more possible candidate of load/store as possible.

day, March 9, 2017 5:41 PM > To: beignet@lists.freedesktop.org > Cc: Yan Wang <yan.w...@linux.intel.com> > Subject: [Beignet] [PATCH v2] Provide more possible candidate of load/store as > possible. > > From: Yan Wang <yan.w...@linux.intel.com> > > Avoid searching r

Re: [Beignet] [PATCH v2] Provide more possible candidate of load/store as possible.

t-boun...@lists.freedesktop.org] On Behalf Of > yan.w...@linux.intel.com > Sent: Thursday, March 9, 2017 5:41 PM > To: beignet@lists.freedesktop.org > Cc: Yan Wang <yan.w...@linux.intel.com> > Subject: [Beignet] [PATCH v2] Provide more possible candidate of load/store as > possib

[Beignet] [PATCH] Provide more possible candidate of load/store as possible.

From: Yan Wang <yan.w...@linux.intel.com> Avoid search range too small in same case like vector of float. It will lead more load/store merged for improving perforamnce. Signed-off-by: Yan Wang <yan.w...@linux.intel.com> --- backend/src/llvm/llvm_loadstore_optimization.cpp | 2 +- 1

[Beignet] [PATCH v2] Provide more possible candidate of load/store as possible.

From: Yan Wang <yan.w...@linux.intel.com> Avoid searching range too small in some case like vector of float. It will lead more load/store merged for improving perforamnce. Signed-off-by: Yan Wang <yan.w...@linux.intel.com> --- backend/src/llvm/llvm_loadstore_optimization.cpp |

[Beignet] [PATCH v2 5/6] Implement TILE_Y large image in clEnqueueReadImage.

From: Yan Wang <yan.w...@linux.intel.com> It will fail to copy data from TILE_Y large image to buffer by memcpy. Use clEnqueueCopyImageToBuffer to do this on GPU side. Signed-off-by: Yan Wang <yan.w...@linux.intel.com> --- src/cl_ap

[Beignet] [PATCH v2 6/6] Implement TILE_Y large image in clEnqueueWriteImage.

From: Yan Wang <yan.w...@linux.intel.com> It will fail to copy data from host ptr to TILE_Y large image by memcpy. Use clEnqueueCopyBufferToImage to do this on GPU side. Signed-off-by: Yan Wang <yan.w...@linux.intel.com> --- src/cl_ap

[Beignet] [PATCH v2 4/6] Implement TILE_Y large image in clEnqueueMapImage and clEnqueueUnmapMemObject.

From: Yan Wang <yan.w...@linux.intel.com> It will fail to copy data from TILE_Y large image to buffer by memcpy. Use clEnqueueCopyImageToBuffer to do this. Signed-off-by: Yan Wang <yan.w...@linux.intel.com> --- src/cl_api_mem.c | 88 +++

[Beignet] [PATCH 4/6] Implement TILE_Y large image in clEnqueueMapImage and clEnqueueUnmapMemObject.

From: Yan Wang <yan.w...@linux.intel.com> It will fail to copy data from TILE_Y large image to buffer by memcpy. Use clEnqueueCopyImageToBuffer to do this. Signed-off-by: Yan Wang <yan.w...@linux.intel.com> --- src/cl_api_mem.c | 88 +++

[Beignet] [PATCH 3/6] Create image with TILE_Y mode still when image size > 128MB for performance.

From: Yan Wang <yan.w...@linux.intel.com> It may failed to copy data from host ptr to TILE_Y large image. So use clCopyBufferToImage to do this on GPU side. Signed-off-by: Yan Wang <yan.w...@linux.intel.com> --- src/cl_mem.c | 100

[Beignet] [PATCH 1/6] Add image copying case for testing large image operations.

From: Yan Wang <yan.w...@linux.intel.com> It is for testing large image with TILE_Y mode. Signed-off-by: Yan Wang <yan.w...@linux.intel.com> --- utests/CMakeLists.txt| 1 + utests/compiler_copy_large_image.cpp | 121 +++ 2 files

[Beignet] [PATCH 2/6] Add image filling case for testing large image operations.

From: Yan Wang <yan.w...@linux.intel.com> It is for testing large image with TILE_Y mode. Signed-off-by: Yan Wang <yan.w...@linux.intel.com> --- utests/CMakeLists.txt| 1 + utests/compiler_fill_large_image.cpp | 124 +++ 2 files

[Beignet] [PATCH 5/6] Implement TILE_Y large image in clEnqueueReadImage.

From: Yan Wang <yan.w...@linux.intel.com> It will fail to copy data from TILE_Y large image to buffer by memcpy. Use clEnqueueCopyImageToBuffer to do this on GPU side. Signed-off-by: Yan Wang <yan.w...@linux.intel.com> --- src/cl_ap

[Beignet] [PATCH 6/6] Implement TILE_Y large image in clEnqueueWriteImage.

From: Yan Wang <yan.w...@linux.intel.com> It will fail to copy data from host ptr to TILE_Y large image by memcpy. Use clEnqueueCopyBufferToImage to do this on GPU side. Signed-off-by: Yan Wang <yan.w...@linux.intel.com> --- src/cl_ap

[Beignet] [PATCH v3 1/8] Add image copying case for testing large image operations.

From: Yan Wang <yan.w...@linux.intel.com> It is for testing large image with TILE_Y mode. Signed-off-by: Yan Wang <yan.w...@linux.intel.com> --- utests/CMakeLists.txt| 1 + utests/compiler_copy_large_image.cpp | 121 +++ 2 files

[Beignet] [PATCH v3 2/8] Add image filling case for testing large image operations.

From: Yan Wang <yan.w...@linux.intel.com> It is for testing large image with TILE_Y mode. Signed-off-by: Yan Wang <yan.w...@linux.intel.com> --- utests/CMakeLists.txt| 1 + utests/compiler_fill_large_image.cpp | 120 +++ 2 files

[Beignet] [PATCH v3 6/8] Implement TILE_Y large image in clEnqueueMapImage and clEnqueueUnmapMemObject.

From: Yan Wang <yan.w...@linux.intel.com> It will fail to copy data from TILE_Y large image to buffer by memcpy. Use clEnqueueCopyImageToBuffer to do this. Signed-off-by: Yan Wang <yan.w...@linux.intel.com> --- src/cl_api_mem.c | 111 ++

[Beignet] [PATCH v3 7/8] Implement TILE_Y large image in clEnqueueReadImage.

From: Yan Wang <yan.w...@linux.intel.com> It will fail to copy data from TILE_Y large image to buffer by memcpy. Use clEnqueueCopyImageToBuffer to do this on GPU side. Signed-off-by: Yan Wang <yan.w...@linux.intel.com> --- src/cl_ap

[Beignet] [PATCH v3 5/8] Create image with TILE_Y mode still when image size>128MB for performance.

From: Yan Wang <yan.w...@linux.intel.com> It may failed to copy data from host ptr to TILE_Y large image. So use clCopyBufferToImage to do this on GPU side. Signed-off-by: Yan Wang <yan.w...@linux.intel.com> --- src/cl_context.c | 6 src/cl_context.h | 2 +- src/cl_mem

[Beignet] [PATCH v3 4/8] Add image use_hostptr benchmark case for testing large image operations.

From: Yan Wang <yan.w...@linux.intel.com> It is for testing large image with TILE_Y mode. Signed-off-by: Yan Wang <yan.w...@linux.intel.com> --- benchmark/CMakeLists.txt | 1 + benchmark/benchmark_use_host_ptr_large_image.cpp | 84 +

[Beignet] [PATCH v3 8/8] Implement TILE_Y large image in clEnqueueWriteImage.

From: Yan Wang <yan.w...@linux.intel.com> It will fail to copy data from host ptr to TILE_Y large image by memcpy. Use clEnqueueCopyBufferToImage to do this on GPU side. Signed-off-by: Yan Wang <yan.w...@linux.intel.com> --- src/cl_ap

[Beignet] [PATCH] Add aligned copy kernels into built-in kernel list.

From: Yan Wang <yan.w...@linux.intel.com> Signed-off-by: Yan Wang <yan.w...@linux.intel.com> --- src/cl_gt_device.h | 8 1 file changed, 8 insertions(+) diff --git a/src/cl_gt_device.h b/src/cl_gt_device.h index f6cb5f8..8008606 100644 --- a/src/cl_gt_device.h +++ b/src/cl

[Beignet] [PATCH v2] Add missed kernel names into built-in kernel list.

From: Yan Wang <yan.w...@linux.intel.com> Signed-off-by: Yan Wang <yan.w...@linux.intel.com> --- src/cl_gt_device.h | 17 + 1 file changed, 17 insertions(+) diff --git a/src/cl_gt_device.h b/src/cl_gt_device.h index f6cb5f8..ff23b32 100644 --- a/src/cl_gt_device

Re: [Beignet] [PATCH v2] Add missed kernel names into built-in kernel list.

m > Sent: Thursday, June 22, 2017 13:52 > To: beignet@lists.freedesktop.org > Cc: Yan Wang <yan.w...@linux.intel.com> > Subject: [Beignet] [PATCH v2] Add missed kernel names into built-in kernel > list. > > From: Yan Wang <yan.w...@linux.intel.com> > > Signe

Re: [Beignet] [PATCH] Add aligned copy kernels into built-in kernel list.

? > -Original Message- > From: Beignet [mailto:beignet-boun...@lists.freedesktop.org] On Behalf Of > yan.w...@linux.intel.com > Sent: Wednesday, June 21, 2017 11:26 > To: beignet@lists.freedesktop.org > Cc: Yan Wang <yan.w...@linux.intel.com> > Subject: [Beignet] [PATC

[Beignet] [PATCH 1/2] Add utest to reproduce the bug of imagedim_non_pow_2 cases of conformance test.

From: Yan Wang <yan.w...@linux.intel.com> Signed-off-by: Yan Wang <yan.w...@linux.intel.com> --- utests/compiler_fill_large_image.cpp | 46 1 file changed, 46 insertions(+) diff --git a/utests/compiler_fill_large_image.c

[Beignet] [PATCH 2/2] Fix bug of clEnqueueCopyBufferToImage and clEnqueueCopyImageToBuffer.

From: Yan Wang <yan.w...@linux.intel.com> "imagedim_non_pow_2" cases of basic modudle of confrmance shows regression after use TILE_Y mode for large image by previous patch. This bug comes from the non-align16 kernel of clEnqueueCopyBufferToImage and clEnqueueCopyImageToBuffe

[Beignet] [PATCH v2 2/2] Fix bug of clEnqueueCopyBufferToImage and clEnqueueCopyImageToBuffer.

From: Yan Wang <yan.w...@linux.intel.com> "imagedim_non_pow_2" cases of basic modudle of confrmance shows regression after use TILE_Y mode for large image by previous patch. This bug comes from the non-align16 kernel of clEnqueueCopyBufferToImage and clEnqueueCopyImageToBuffe

[Beignet] [PATCH v2 2/2] Fix bug of size of tmp_ker_buf for TILE_Y copying of large image.

From: Yan Wang <yan.w...@linux.intel.com> 1. The size should be calculated based region and bpp of image instead of the whole image size. 2. When use blocking mode, the copying kernel need be finished. Otherwise, it will cause allocations of conformance test failed. Signed-off-by: Ya

[Beignet] [PATCH 2/2] Fix bug of size of tmp_ker_buf for TILE_Y copying of large image.

From: Yan Wang <yan.w...@linux.intel.com> the size should be calculated based region and bpp of image instead of the whole image size. Otherwise, it will cause allocations of conformance test failed. Signed-off-by: Yan Wang <yan.w...@linux.intel.com> --- src/cl_api_mem.c | 15 ++

[Beignet] [PATCH 1/2] Add utest case for filling image by small region.

From: Yan Wang <yan.w...@linux.intel.com> It is used to reproduce the bug of allocations of conformance test. Signed-off-by: Yan Wang <yan.w...@linux.intel.com> --- utests/compiler_fill_large_image.cpp | 50 1 file changed, 50 insertions(+)

[Beignet] [PATCH v3 7/7] Optimize clEnqueueWriteImageByKernel and clEnqueuReadImageByKernel.

From: Yan Wang <yan.w...@linux.intel.com> 1. Only copy the data by origin and region defined. 2. Add clFinish to guarantee the kernel copying is finished when blocking writing. Signed-off-by: Yan Wang <yan.w...@linux.intel.com> --- src/cl_api_mem.c | 15 +++ 1 file

[Beignet] [PATCH v3 4/7] Add cl_mem_record_map_mem_for_kernel() for record map adress for TILE_Y image by kernel copying.

From: Yan Wang <yan.w...@linux.intel.com> Signed-off-by: Yan Wang <yan.w...@linux.intel.com> --- src/cl_mem.c | 109 +-- src/cl_mem.h | 5 +++ 2 files changed, 88 insertions(+), 26 deletions(-) diff --git a/src/cl_mem.c b/

[Beignet] [PATCH v3 3/7] Add utest to test writing data into large image (TILE_Y) by map/unmap and USE_HOST_PTR mode.

From: Yan Wang <yan.w...@linux.intel.com> Signed-off-by: Yan Wang <yan.w...@linux.intel.com> --- utests/runtime_use_host_ptr_large_image.cpp | 109 1 file changed, 109 insertions(+) diff --git a/utests/runtime_use_host_ptr_large_image.c

  1   2   >