From: Pan Xiuli
Add sub group broadcast and intel sub group broadcast for short type.
Signed-off-by: Pan Xiuli
---
backend/src/libocl/tmpl/ocl_simd.tmpl.cl | 9 +
backend/src/libocl/tmpl/ocl_simd.tmpl.h | 4
2 files changed, 13
From: Pan Xiuli
Add intel subgroup short mem bleck read/write and image block read/write
also fix some old block read/write bug.
Refine old uint block read/write with _ui suffix.
Signed-off-by: Pan Xiuli
---
backend/src/backend/gen_context.cpp
From: Pan Xiuli
Check if the device support intel_subgroups_short extension, also check
if the device support intel_subgroups extension first.
Signed-off-by: Pan Xiuli
---
utests/utest_helper.cpp | 20
utests/utest_helper.hpp | 2
From: Pan Xiuli
We support intel_subgroups_short extension.
Signed-off-by: Pan Xiuli
---
backend/src/libocl/include/ocl.h | 1 +
src/cl_extensions.h | 5 +++--
2 files changed, 4 insertions(+), 2 deletions(-)
diff --git
From: Pan Xiuli
Convert the subnr into nr if the subnr is bigger than 32, this will make
offset usage universally.
Signed-off-by: Pan Xiuli
---
backend/src/backend/gen_register.hpp | 4
1 file changed, 4 insertions(+)
diff --git
From: Pan Xiuli
Signed-off-by: Pan Xiuli
---
kernels/compiler_sub_group_shuffle.cl | 22 +++-
kernels/compiler_sub_group_shuffle_down.cl | 23 -
kernels/compiler_sub_group_shuffle_up.cl | 23 -
From: Pan Xiuli
Add ushort block read/write for buffer and image.
Refine uint block read/write with suffix _ui.
Signed-off-by: Pan Xiuli
---
kernels/compiler_subgroup_buffer_block_read.cl | 47 ++--
From: Pan Xiuli
Signed-off-by: Pan Xiuli
---
kernels/compiler_subgroup_broadcast.cl | 10 ++
utests/compiler_subgroup_broadcast.cpp | 11 +++
2 files changed, 21 insertions(+)
diff --git a/kernels/compiler_subgroup_broadcast.cl
From: Pan Xiuli
Add short type sub group shuffle(simd shuffle)
Signed-off-by: Pan Xiuli
---
backend/src/ir/instruction.cpp | 5 +++--
backend/src/libocl/tmpl/ocl_simd.tmpl.cl | 6 ++
backend/src/libocl/tmpl/ocl_simd.tmpl.h | 11
From: Pan Xiuli
Simd shuffle should support different type, we used to support float or
dword type. Now we can set offset by src type.
Signed-off-by: Pan Xiuli
---
backend/src/backend/gen_context.cpp | 2 +-
1 file changed, 1 insertion(+), 1
From: Pan Xiuli
Signed-off-by: Pan Xiuli
---
kernels/compiler_subgroup_reduce.cl | 22 ++
kernels/compiler_subgroup_scan_exclusive.cl | 36
kernels/compiler_subgroup_scan_inclusive.cl | 36
From: Pan Xiuli
To unpack UW we may need to add mov and we do not want this mov to be
optimizated by the sel ir optimization. Add check for hstrid to avoid
this kind optimization.
Signed-off-by: Pan Xiuli
---
From: Pan Xiuli
Add intel sub group short type builtins.
Signed-off-by: Pan Xiuli
---
backend/src/backend/gen_context.cpp | 12 +
backend/src/libocl/tmpl/ocl_simd.tmpl.cl | 44
Hi Rebecca,
This version LGTM except some points need to be minor refined. Just see
my comments below.
-Original Message-
From: Beignet [mailto:beignet-boun...@lists.freedesktop.org] On Behalf Of
Rebecca N. Palmer
Sent: Wednesday, October 12, 2016 5:50 AM
To: Weng, Chuanbo
2baff9c moved mem->magic to cl_base_object.
---
(Or should this be CL_OBJECT_IS_MEM(mem), i.e. also checking the reference
count?)
--- a/src/cl_cmrt.cpp
+++ b/src/cl_cmrt.cpp
@@ -256,7 +256,7 @@ cl_int cmrt_set_kernel_arg(cl_kernel k,
result = cmrt_kernel->SetKernelArg(index, sz, value);
LGTM, thanks.
-Original Message-
From: Beignet [mailto:beignet-boun...@lists.freedesktop.org] On Behalf Of
Rebecca N. Palmer
Sent: Thursday, October 13, 2016 6:15 AM
To: beignet@lists.freedesktop.org
Subject: [Beignet] [PATCH] Fix build failure with CMRT enabled
2baff9c moved mem->magic
16 matches
Mail list logo