On 2015-05-28 13:04, Grigori Goronzy wrote:
Work-group size should always be aligned to subgroup size; this is a
basic requirement, otherwise some work-items will be no-operation.
It might make sense to refine the value according to a kernel's
resource usage, but that's a possible optimization for the future.
Ping?
This is rather simple, but I'd like an Rb, if possible. That also goes
for the Gallium support patch.
Grigori
---
src/gallium/state_trackers/clover/api/kernel.cpp | 2 +-
src/gallium/state_trackers/clover/core/device.cpp | 5 +++++
src/gallium/state_trackers/clover/core/device.hpp | 1 +
3 files changed, 7 insertions(+), 1 deletion(-)
diff --git a/src/gallium/state_trackers/clover/api/kernel.cpp
b/src/gallium/state_trackers/clover/api/kernel.cpp
index 05cc392..857a152 100644
--- a/src/gallium/state_trackers/clover/api/kernel.cpp
+++ b/src/gallium/state_trackers/clover/api/kernel.cpp
@@ -169,7 +169,7 @@ clGetKernelWorkGroupInfo(cl_kernel d_kern,
cl_device_id d_dev,
break;
case CL_KERNEL_PREFERRED_WORK_GROUP_SIZE_MULTIPLE:
- buf.as_scalar<size_t>() = 1;
+ buf.as_scalar<size_t>() = dev.subgroup_size();
break;
case CL_KERNEL_PRIVATE_MEM_SIZE:
diff --git a/src/gallium/state_trackers/clover/core/device.cpp
b/src/gallium/state_trackers/clover/core/device.cpp
index 42b45b7..c42d1d2 100644
--- a/src/gallium/state_trackers/clover/core/device.cpp
+++ b/src/gallium/state_trackers/clover/core/device.cpp
@@ -185,6 +185,11 @@ device::max_block_size() const {
return { v.begin(), v.end() };
}
+cl_uint
+device::subgroup_size() const {
+ return get_compute_param<uint32_t>(pipe,
PIPE_COMPUTE_CAP_SUBGROUP_SIZE)[0];
+}
+
std::string
device::device_name() const {
return pipe->get_name(pipe);
diff --git a/src/gallium/state_trackers/clover/core/device.hpp
b/src/gallium/state_trackers/clover/core/device.hpp
index de5fc6b..2857847 100644
--- a/src/gallium/state_trackers/clover/core/device.hpp
+++ b/src/gallium/state_trackers/clover/core/device.hpp
@@ -67,6 +67,7 @@ namespace clover {
bool has_doubles() const;
std::vector<size_t> max_block_size() const;
+ cl_uint subgroup_size() const;
std::string device_name() const;
std::string vendor_name() const;
enum pipe_shader_ir ir_format() const;
_______________________________________________
mesa-dev mailing list
mesa-dev@lists.freedesktop.org
http://lists.freedesktop.org/mailman/listinfo/mesa-dev
