[Beignet] [PATCH 6/9] runtime: fix clLinkProgram bug.

From: Luo Xionghu clLinkProgram need check the existence of "-cl-kernel-arg-info" build_option of all the input_programs. User may link two SPIR program and call clGetKernelArgInfo to query kernel args. Signed-off-by: Luo Xionghu ---

[Beignet] [PATCH 8/9] gbe: add vec_type_hint's type into functionAttributes.

From: Luo Xionghu for SPIR kernel, user may call clGetKernelInfo with CL_KERNEL_ATTRIBUTES to query the functionAttributes. Signed-off-by: Luo Xionghu --- backend/src/ir/type.cpp | 6 ++-- backend/src/llvm/llvm_gen_backend.cpp | 65

[Beignet] [PATCH 7/9] runtime: add missing supported format image_1d_buffer.

From: Luo Xionghu Signed-off-by: Luo Xionghu --- src/cl_api.c | 1 + 1 file changed, 1 insertion(+) diff --git a/src/cl_api.c b/src/cl_api.c index e5ee002..a56b596 100644 --- a/src/cl_api.c +++ b/src/cl_api.c @@ -682,6 +682,7 @@

[Beignet] [PATCH 3/9] gbe/libocl: change xxx_fence function to OVERLOADABLE.

From: Luo Xionghu the mem_fence, read_mem_fence and write_mem_fence functions need be OVERLOADABLE for link. Signed-off-by: Luo Xionghu --- backend/src/libocl/include/ocl_sync.h | 6 +++--- backend/src/libocl/src/ocl_sync.cl| 6 +++--- 2 files

[Beignet] [PATCH 9/9] gbe/libocl: define the gentype half_xxx math function instead of using MACRO.

From: Luo Xionghu these function need to be overloadable for link. Signed-off-by: Luo Xionghu --- backend/src/libocl/script/ocl_math.def | 28 ++--- backend/src/libocl/tmpl/ocl_math.tmpl.cl | 42

[Beignet] [PATCH 1/9] gbe/libocl: define the vloada_xxx function instead of using MACRO.

From: Luo Xionghu these function need to be overloadable for link. Signed-off-by: Luo Xionghu --- backend/src/libocl/include/ocl_vload.h | 13 ++--- backend/src/libocl/src/ocl_vload.cl| 21 - 2 files changed, 26

[Beignet] [PATCH 5/9] runtime: fix clCompileProgram bug.

From: Luo Xionghu forgot to add FROM_LLVM_SPIR in compileProgram; the BINARY_TYPE is BINARY_TYPE_INTERMIDIATE if create from SPIR binary. Signed-off-by: Luo Xionghu --- src/cl_api.c | 1 + src/cl_program.c | 5 - src/cl_program.h | 1 + 3

[Beignet] [PATCH 4/9] runtime: initialize the memory content to 0.

From: Luo Xionghu Signed-off-by: Luo Xionghu --- src/cl_api.c | 1 + 1 file changed, 1 insertion(+) diff --git a/src/cl_api.c b/src/cl_api.c index ddd39cf..9eb4ffd 100644 --- a/src/cl_api.c +++ b/src/cl_api.c @@ -56,6 +56,7 @@ typedef intptr_t

[Beignet] [PATCH 2/9] gbe: use kernel_arg_base_type to recognize image arguments.

From: Luo Xionghu we use string compare function to recognize the image type, but image types could be defined to other type, so use kernel_arg_base_type to recognize image arguments. --- backend/src/ir/function.hpp | 15 ---

[Beignet] [PATCH 2/5] GBE: don't assert even if we fail to compile kernel at the backend stage.

We should not assert even if the application triggers a internal limitation such as lack of scratch space. We should return error to the application and let the application to make further decision. Signed-off-by: Zhigang Gong --- backend/src/backend/context.cpp

[Beignet] [PATCH 4/5] runtime: set CL_KERNEL_PREFERRED_WORK_GROUP_SIZE_MULTIPLE to kernel's SIMD_WIDTH.

It makes sense to set CL_KERNEL_PREFERRED_WORK_GROUP_SIZE_MULTIPLE to the corresponding SIMD size. Then it provides a way for intel's OCL application to get SIMD width at runtime and make some SIMD width dependant optimization possible. Signed-off-by: Zhigang Gong ---

[Beignet] [PATCH 3/5] GBE: remove useless assertions code.

Signed-off-by: Zhigang Gong --- backend/src/backend/context.cpp | 14 +- 1 file changed, 5 insertions(+), 9 deletions(-) diff --git a/backend/src/backend/context.cpp b/backend/src/backend/context.cpp index 47d8a45..5f5a858 100644 ---

[Beignet] [PATCH 1/5] GBE: extent register allocator size/offset to 32bit.

Because the range of scratch size exceed the int16_t's maximum size. We have to extent these elements to 32 bit. Signed-off-by: Zhigang Gong --- backend/src/backend/context.cpp | 52 - backend/src/backend/context.hpp | 6 ++--- 2

[Beignet] [PATCH 5/5] GBE: decrease the loop unrolling threshold to 640.

1024 is some how too large for some kernels and may cause some kernels fail to build due to lack of enough scratching space. Signed-off-by: Zhigang Gong --- backend/src/llvm/llvm_to_gen.cpp | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git

[Beignet] [PATCH] runtime: silent some error messages.

We already set corresponding error code and return it to the caller. Don't bother to print the error messages in beignet internal. Signed-off-by: Zhigang Gong --- src/cl_command_queue_gen7.c | 3 --- 1 file changed, 3 deletions(-) diff --git