Merge pull request #169 from Devsh-Graphics-Programming/ali_blur3

initialize blur example

Author: devshgraphicsprogramming

26_Blur/CMakeLists.txt

@@ -0,0 +1,24 @@
+include(common RESULT_VARIABLE RES)
+if(NOT RES)
+message(FATAL_ERROR "common.cmake not found. Should be in {repo_root}/cmake directory")
+endif()
+
+nbl_create_executable_project("" "" "" "" "${NBL_EXECUTABLE_PROJECT_CREATION_PCH_TARGET}")
+
+if(NBL_EMBED_BUILTIN_RESOURCES)
+set(_BR_TARGET_ ${EXECUTABLE_NAME}_builtinResourceData)
+set(RESOURCE_DIR "app_resources")
+
+get_filename_component(_SEARCH_DIRECTORIES_ "${CMAKE_CURRENT_SOURCE_DIR}" ABSOLUTE)
+get_filename_component(_OUTPUT_DIRECTORY_SOURCE_ "${CMAKE_CURRENT_BINARY_DIR}/src" ABSOLUTE)
+get_filename_component(_OUTPUT_DIRECTORY_HEADER_ "${CMAKE_CURRENT_BINARY_DIR}/include" ABSOLUTE)
+
+ file(GLOB_RECURSE BUILTIN_RESOURCE_FILES RELATIVE "${CMAKE_CURRENT_SOURCE_DIR}/${RESOURCE_DIR}" CONFIGURE_DEPENDS "${CMAKE_CURRENT_SOURCE_DIR}/${RESOURCE_DIR}/*")
+ foreach(RES_FILE ${BUILTIN_RESOURCE_FILES})
+ LIST_BUILTIN_RESOURCE(RESOURCES_TO_EMBED "${RES_FILE}")
+ endforeach()
+
+ADD_CUSTOM_BUILTIN_RESOURCES(${_BR_TARGET_} RESOURCES_TO_EMBED "${_SEARCH_DIRECTORIES_}" "${RESOURCE_DIR}" "nbl::this_example::builtin" "${_OUTPUT_DIRECTORY_HEADER_}" "${_OUTPUT_DIRECTORY_SOURCE_}")
+
+LINK_BUILTIN_RESOURCES_TO_TARGET(${EXECUTABLE_NAME} ${_BR_TARGET_})
+endif()

26_Blur/app_resources/common.hlsl

@@ -0,0 +1,11 @@
+#include "nbl/builtin/hlsl/cpp_compat.hlsl"
+#include "nbl/builtin/hlsl/type_traits.hlsl"
+
+static const uint16_t PASSES = 2;
+
+struct PushConstants
+{
+ nbl::hlsl::float32_t radius;
+uint32_t activeAxis : 2;
+ uint32_t edgeWrapMode : 6;
+};

26_Blur/app_resources/shader.comp.hlsl

@@ -0,0 +1,162 @@
+#include "nbl/builtin/hlsl/prefix_sum_blur/blur.hlsl"
+#include "nbl/builtin/hlsl/prefix_sum_blur/box_sampler.hlsl"
+#include "nbl/builtin/hlsl/workgroup/scratch_size.hlsl"
+#include "nbl/builtin/hlsl/jit/device_capabilities.hlsl"
+#include "nbl/builtin/hlsl/colorspace/OETF.hlsl"
+#include "common.hlsl"
+
+using namespace nbl::hlsl;
+
+uint32_t3 glsl::gl_WorkGroupSize() { return uint32_t3(WORKGROUP_SIZE, 1, 1); }
+
+[[vk::binding(0)]]
+Texture2D<float32_t4> input;
+[[vk::binding(1)]]
+RWTexture2D<float32_t4> output;
+
+[[vk::push_constant]] PushConstants pc;
+
+template<uint16_t Chnls>
+struct TextureProxy
+{
+ NBL_CONSTEXPR uint16_t Channels = Chnls;
+ using texel_t = vector<float32_t, Channels>;
+
+ // divisions by PoT constant will optimize out nicely
+ template<typename T>
+ T get(const uint16_t channel, const uint16_t uv)
+ {
+ return spill[uv / WORKGROUP_SIZE][channel];
+ }
+
+ template<typename T>
+void set(const uint16_t channel, const uint16_t uv, T value)
+{
+ spill[uv / WORKGROUP_SIZE][channel] = value;
+}
+
+ void load()
+ {
+ const uint16_t end = linearSize();
+ uint16_t ix = workgroup::SubgroupContiguousIndex();
+ // because workgroups do scans cooperatively all spill values need sane defaults
+ for (uint16_t i=0; i < SpillSize; ix += WORKGROUP_SIZE)
+ spill[i++] = ix < end ? (texel_t)input[position(ix)] : promote<texel_t>(0.f);
+ }
+
+ void store()
+ {
+ const uint16_t end = linearSize();
+ uint16_t i = 0;
+ // making sure that we don't store out of range
+ for (uint16_t ix = workgroup::SubgroupContiguousIndex(); ix < end; ix += WORKGROUP_SIZE)
+ {
+ float32_t4 tmp = float32_t4(0, 0, 0, 1);
+ for (uint16_t ch=0; ch < Channels; ch++)
+ tmp[ch] = spill[i][ch];
+ i++;
+ output[position(ix)] = tmp;
+ }
+ }
+
+ uint16_t linearSize()
+ {
+ uint32_t3 dims;
+ input.GetDimensions(0, dims.x, dims.y, dims.z);
+ return _static_cast<uint16_t>(dims[activeAxis]);
+ }
+
+ uint16_t2 position(uint16_t ix)
+ {
+ uint16_t2 pos;
+ pos[activeAxis] = ix;
+ pos[activeAxis ^ 0x1] = _static_cast<uint16_t>(glsl::gl_WorkGroupID().x);
+ return pos;
+ }
+
+ // whether we pas along X or Y
+ uint16_t activeAxis;
+ NBL_CONSTEXPR uint16_t SpillSize = (MAX_SCANLINE_SIZE - 1) / WORKGROUP_SIZE + 1;
+ texel_t spill[SpillSize];
+};
+
+static const uint16_t MAX_SCAN_SCRATCH_SIZE = workgroup::scratch_size_arithmetic<WORKGROUP_SIZE, MAX_SUBGROUP_SIZE>::value + 2;
+
+// we always use `uint32_t`
+groupshared uint32_t smem[MAX_SCANLINE_SIZE];
+groupshared uint32_t prefix_smem[MAX_SCAN_SCRATCH_SIZE];
+
+struct SharedMemoryProxy
+{
+ NBL_CONSTEXPR uint16_t Size = MAX_SCANLINE_SIZE;
+
+ template<typename T, typename I = uint16_t>
+enable_if_t<sizeof(T) == sizeof(uint32_t), T> get(const I idx)
+{
+return bit_cast<T>(smem[idx]);
+}
+
+template<typename T, typename I = uint16_t>
+enable_if_t<sizeof(T) == sizeof(uint32_t), void> set(const I idx, T value)
+{
+ smem[idx] = bit_cast<uint32_t>(value);
+}
+
+ void workgroupExecutionAndMemoryBarrier()
+ {
+ glsl::barrier();
+ }
+};
+
+struct ScanSharedMemoryProxy
+{
+ NBL_CONSTEXPR uint16_t Size = MAX_SCAN_SCRATCH_SIZE;
+
+ // these get used by Box1D
+ template<typename T, typename I = uint16_t>
+enable_if_t<sizeof(T) == sizeof(uint32_t), T> get(const I idx)
+{
+return bit_cast<T>(prefix_smem[idx]);
+}
+
+template<typename T, typename I = uint16_t>
+enable_if_t<sizeof(T) == sizeof(uint32_t), void> set(const I idx, T value)
+{
+ prefix_smem[idx] = bit_cast<uint32_t>(value);
+}
+
+ void workgroupExecutionAndMemoryBarrier()
+ {
+ glsl::barrier();
+ }
+};
+
+[numthreads(WORKGROUP_SIZE, 1, 1)]
+void main()
+{
+ ScanSharedMemoryProxy scanSmemAccessor;
+
+ TextureProxy<CHANNELS> texAccessor;
+ texAccessor.activeAxis = (uint16_t)pc.activeAxis;
+ texAccessor.load();
+ 
+ prefix_sum_blur::BoxSampler<SharedMemoryProxy, float32_t> boxSampler;
+ boxSampler.wrapMode = uint16_t(pc.edgeWrapMode);
+ boxSampler.linearSize = texAccessor.linearSize();
+ boxSampler.normalizationFactor = 1 / (2 * pc.radius + 1);
+
+ prefix_sum_blur::Blur1D<decltype(texAccessor), decltype(scanSmemAccessor), decltype(boxSampler), WORKGROUP_SIZE, jit::device_capabilities> blur;
+ blur.radius = pc.radius;
+ blur.borderColor = float32_t4(0, 1, 0, 1);
+
+ for (uint16_t ch=0; ch < CHANNELS; ch++)
+ for (uint16_t pass=0; pass < PASSES; pass++)
+ {
+ // its the `SharedMemoryProxy` that gets aliased and reused so we need to barrier on its memory
+ if (ch != 0 && pass != 0)
+ boxSampler.prefixSumAccessor.workgroupExecutionAndMemoryBarrier();
+ blur(texAccessor, scanSmemAccessor, boxSampler, ch);
+ }
+
+ texAccessor.store();
+}