Unverified Commit 016ebaa7 authored by carlushuang's avatar carlushuang Committed by GitHub
Browse files

support dynamic buffer using memory coherence glc_slc bit from template (#725)

parent 1dd455d6
...@@ -19,7 +19,8 @@ namespace ck { ...@@ -19,7 +19,8 @@ namespace ck {
template <AddressSpaceEnum BufferAddressSpace, template <AddressSpaceEnum BufferAddressSpace,
typename T, typename T,
typename ElementSpaceSize, typename ElementSpaceSize,
bool InvalidElementUseNumericalZeroValue> bool InvalidElementUseNumericalZeroValue,
AmdBufferCoherenceEnum coherence = AmdBufferCoherenceEnum::DefaultCoherence>
struct DynamicBuffer struct DynamicBuffer
{ {
using type = T; using type = T;
...@@ -77,13 +78,16 @@ struct DynamicBuffer ...@@ -77,13 +78,16 @@ struct DynamicBuffer
if constexpr(InvalidElementUseNumericalZeroValue) if constexpr(InvalidElementUseNumericalZeroValue)
{ {
return amd_buffer_load_invalid_element_return_zero<remove_cvref_t<T>, t_per_x>( return amd_buffer_load_invalid_element_return_zero<remove_cvref_t<T>,
t_per_x,
coherence>(
p_data_, i, is_valid_element, element_space_size_); p_data_, i, is_valid_element, element_space_size_);
} }
else else
{ {
return amd_buffer_load_invalid_element_return_customized_value<remove_cvref_t<T>, return amd_buffer_load_invalid_element_return_customized_value<remove_cvref_t<T>,
t_per_x>( t_per_x,
coherence>(
p_data_, i, is_valid_element, element_space_size_, invalid_element_value_); p_data_, i, is_valid_element, element_space_size_, invalid_element_value_);
} }
} }
...@@ -173,7 +177,7 @@ struct DynamicBuffer ...@@ -173,7 +177,7 @@ struct DynamicBuffer
{ {
constexpr index_t t_per_x = scalar_per_x_vector / scalar_per_t_vector; constexpr index_t t_per_x = scalar_per_x_vector / scalar_per_t_vector;
amd_buffer_store<remove_cvref_t<T>, t_per_x>( amd_buffer_store<remove_cvref_t<T>, t_per_x, coherence>(
x, p_data_, i, is_valid_element, element_space_size_); x, p_data_, i, is_valid_element, element_space_size_);
} }
else if constexpr(GetAddressSpace() == AddressSpaceEnum::Lds && else if constexpr(GetAddressSpace() == AddressSpaceEnum::Lds &&
...@@ -376,14 +380,19 @@ struct DynamicBuffer ...@@ -376,14 +380,19 @@ struct DynamicBuffer
__host__ __device__ static constexpr bool IsDynamicBuffer() { return true; } __host__ __device__ static constexpr bool IsDynamicBuffer() { return true; }
}; };
template <AddressSpaceEnum BufferAddressSpace, typename T, typename ElementSpaceSize> template <AddressSpaceEnum BufferAddressSpace,
AmdBufferCoherenceEnum coherence = AmdBufferCoherenceEnum::DefaultCoherence,
typename T,
typename ElementSpaceSize>
__host__ __device__ constexpr auto make_dynamic_buffer(T* p, ElementSpaceSize element_space_size) __host__ __device__ constexpr auto make_dynamic_buffer(T* p, ElementSpaceSize element_space_size)
{ {
return DynamicBuffer<BufferAddressSpace, T, ElementSpaceSize, true>{p, element_space_size}; return DynamicBuffer<BufferAddressSpace, T, ElementSpaceSize, true, coherence>{
p, element_space_size};
} }
template < template <
AddressSpaceEnum BufferAddressSpace, AddressSpaceEnum BufferAddressSpace,
AmdBufferCoherenceEnum coherence = AmdBufferCoherenceEnum::DefaultCoherence,
typename T, typename T,
typename ElementSpaceSize, typename ElementSpaceSize,
typename X, typename X,
...@@ -391,7 +400,7 @@ template < ...@@ -391,7 +400,7 @@ template <
__host__ __device__ constexpr auto __host__ __device__ constexpr auto
make_dynamic_buffer(T* p, ElementSpaceSize element_space_size, X invalid_element_value) make_dynamic_buffer(T* p, ElementSpaceSize element_space_size, X invalid_element_value)
{ {
return DynamicBuffer<BufferAddressSpace, T, ElementSpaceSize, false>{ return DynamicBuffer<BufferAddressSpace, T, ElementSpaceSize, false, coherence>{
p, element_space_size, invalid_element_value}; p, element_space_size, invalid_element_value};
} }
......
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment