diff options
Diffstat (limited to 'thirdparty/embree/common/algorithms')
4 files changed, 58 insertions, 18 deletions
diff --git a/thirdparty/embree/common/algorithms/parallel_for.h b/thirdparty/embree/common/algorithms/parallel_for.h index 645681ac63..6d411e4852 100644 --- a/thirdparty/embree/common/algorithms/parallel_for.h +++ b/thirdparty/embree/common/algorithms/parallel_for.h @@ -26,7 +26,6 @@ namespace embree abort(); // -- GODOT end -- } - #elif defined(TASKING_TBB) #if TBB_INTERFACE_VERSION >= 12002 tbb::task_group_context context; diff --git a/thirdparty/embree/common/algorithms/parallel_for_for.h b/thirdparty/embree/common/algorithms/parallel_for_for.h index 92c37a4a38..7838ef11b3 100644 --- a/thirdparty/embree/common/algorithms/parallel_for_for.h +++ b/thirdparty/embree/common/algorithms/parallel_for_for.h @@ -30,15 +30,20 @@ namespace embree template<typename ArrayArray> __forceinline ParallelForForState (ArrayArray& array2, const size_t minStepSize) { init(array2,minStepSize); + } + + template<typename SizeFunc> + __forceinline ParallelForForState (const size_t numArrays, const SizeFunc& getSize, const size_t minStepSize) { + init(numArrays,getSize,minStepSize); } - template<typename ArrayArray> - __forceinline void init ( ArrayArray& array2, const size_t minStepSize ) + template<typename SizeFunc> + __forceinline void init ( const size_t numArrays, const SizeFunc& getSize, const size_t minStepSize ) { /* first calculate total number of elements */ size_t N = 0; - for (size_t i=0; i<array2.size(); i++) { - N += array2[i] ? array2[i]->size() : 0; + for (size_t i=0; i<numArrays; i++) { + N += getSize(i); } this->N = N; @@ -54,8 +59,8 @@ namespace embree size_t k0 = (++taskIndex)*N/taskCount; for (size_t i=0, k=0; taskIndex < taskCount; i++) { - assert(i<array2.size()); - size_t j=0, M = array2[i] ? array2[i]->size() : 0; + assert(i<numArrays); + size_t j=0, M = getSize(i); while (j<M && k+M-j >= k0 && taskIndex < taskCount) { assert(taskIndex<taskCount); i0[taskIndex] = i; @@ -67,6 +72,12 @@ namespace embree } } + template<typename ArrayArray> + __forceinline void init ( ArrayArray& array2, const size_t minStepSize ) + { + init(array2.size(),[&](size_t i) { return array2[i] ? array2[i]->size() : 0; },minStepSize); + } + __forceinline size_t size() const { return N; } diff --git a/thirdparty/embree/common/algorithms/parallel_for_for_prefix_sum.h b/thirdparty/embree/common/algorithms/parallel_for_for_prefix_sum.h index b15b44a991..8c3f4aace7 100644 --- a/thirdparty/embree/common/algorithms/parallel_for_for_prefix_sum.h +++ b/thirdparty/embree/common/algorithms/parallel_for_for_prefix_sum.h @@ -17,15 +17,20 @@ namespace embree __forceinline ParallelForForPrefixSumState (ArrayArray& array2, const size_t minStepSize) : ParallelForForState(array2,minStepSize) {} + template<typename SizeFunc> + __forceinline ParallelForForPrefixSumState (size_t numArrays, const SizeFunc& getSize, const size_t minStepSize) + : ParallelForForState(numArrays,getSize,minStepSize) {} + ParallelPrefixSumState<Value> prefix_state; }; - template<typename ArrayArray, typename Index, typename Value, typename Func, typename Reduction> - __forceinline Value parallel_for_for_prefix_sum0( ParallelForForPrefixSumState<Value>& state, ArrayArray& array2, Index minStepSize, - const Value& identity, const Func& func, const Reduction& reduction) + template<typename SizeFunc, typename Index, typename Value, typename Func, typename Reduction> + __forceinline Value parallel_for_for_prefix_sum0_( ParallelForForPrefixSumState<Value>& state, Index minStepSize, + const SizeFunc& getSize, const Value& identity, const Func& func, const Reduction& reduction) { /* calculate number of tasks to use */ const size_t taskCount = state.taskCount; + /* perform parallel prefix sum */ parallel_for(taskCount, [&](const size_t taskIndex) { @@ -38,9 +43,9 @@ namespace embree size_t k=k0; Value N=identity; for (size_t i=i0; k<k1; i++) { - const size_t size = array2[i] ? array2[i]->size() : 0; + const size_t size = getSize(i); const size_t r0 = j0, r1 = min(size,r0+k1-k); - if (r1 > r0) N = reduction(N, func(array2[i],range<Index>((Index)r0,(Index)r1),(Index)k,(Index)i)); + if (r1 > r0) N = reduction(N, func((Index)i,range<Index>((Index)r0,(Index)r1),(Index)k)); k+=r1-r0; j0 = 0; } state.prefix_state.counts[taskIndex] = N; @@ -58,9 +63,10 @@ namespace embree return sum; } - template<typename ArrayArray, typename Index, typename Value, typename Func, typename Reduction> - __forceinline Value parallel_for_for_prefix_sum1( ParallelForForPrefixSumState<Value>& state, ArrayArray& array2, Index minStepSize, - const Value& identity, const Func& func, const Reduction& reduction) + template<typename SizeFunc, typename Index, typename Value, typename Func, typename Reduction> + __forceinline Value parallel_for_for_prefix_sum1_( ParallelForForPrefixSumState<Value>& state, Index minStepSize, + const SizeFunc& getSize, + const Value& identity, const Func& func, const Reduction& reduction) { /* calculate number of tasks to use */ const size_t taskCount = state.taskCount; @@ -76,9 +82,9 @@ namespace embree size_t k=k0; Value N=identity; for (size_t i=i0; k<k1; i++) { - const size_t size = array2[i] ? array2[i]->size() : 0; + const size_t size = getSize(i); const size_t r0 = j0, r1 = min(size,r0+k1-k); - if (r1 > r0) N = reduction(N, func(array2[i],range<Index>((Index)r0,(Index)r1),(Index)k,(Index)i,reduction(state.prefix_state.sums[taskIndex],N))); + if (r1 > r0) N = reduction(N, func((Index)i,range<Index>((Index)r0,(Index)r1),(Index)k,reduction(state.prefix_state.sums[taskIndex],N))); k+=r1-r0; j0 = 0; } state.prefix_state.counts[taskIndex] = N; @@ -96,6 +102,30 @@ namespace embree return sum; } + template<typename ArrayArray, typename Index, typename Value, typename Func, typename Reduction> + __forceinline Value parallel_for_for_prefix_sum0( ParallelForForPrefixSumState<Value>& state, + ArrayArray& array2, Index minStepSize, + const Value& identity, const Func& func, const Reduction& reduction) + { + return parallel_for_for_prefix_sum0_(state,minStepSize, + [&](Index i) { return array2[i] ? array2[i]->size() : 0; }, + identity, + [&](Index i, const range<Index>& r, Index k) { return func(array2[i], r, k, i); }, + reduction); + } + + template<typename ArrayArray, typename Index, typename Value, typename Func, typename Reduction> + __forceinline Value parallel_for_for_prefix_sum1( ParallelForForPrefixSumState<Value>& state, + ArrayArray& array2, Index minStepSize, + const Value& identity, const Func& func, const Reduction& reduction) + { + return parallel_for_for_prefix_sum1_(state,minStepSize, + [&](Index i) { return array2[i] ? array2[i]->size() : 0; }, + identity, + [&](Index i, const range<Index>& r, Index k, const Value& base) { return func(array2[i], r, k, i, base); }, + reduction); + } + template<typename ArrayArray, typename Value, typename Func, typename Reduction> __forceinline Value parallel_for_for_prefix_sum0( ParallelForForPrefixSumState<Value>& state, ArrayArray& array2, const Value& identity, const Func& func, const Reduction& reduction) diff --git a/thirdparty/embree/common/algorithms/parallel_reduce.h b/thirdparty/embree/common/algorithms/parallel_reduce.h index 8271372ea4..cd0078f2e6 100644 --- a/thirdparty/embree/common/algorithms/parallel_reduce.h +++ b/thirdparty/embree/common/algorithms/parallel_reduce.h @@ -26,7 +26,7 @@ namespace embree const Index threadCount = (Index) TaskScheduler::threadCount(); taskCount = min(taskCount,threadCount,maxTasks); - /* parallel invokation of all tasks */ + /* parallel invocation of all tasks */ dynamic_large_stack_array(Value,values,taskCount,8192); // consumes at most 8192 bytes on the stack parallel_for(taskCount, [&](const Index taskIndex) { const Index k0 = first+(taskIndex+0)*(last-first)/taskCount; |