copy examples to CE container in the build time

tools/nsc/CMakeLists.txt

@@ -166,6 +166,7 @@ endif()
 
 make_directory("${NBL_DOCKER_CTX_DIR}/Runtimes")
 make_directory("${NBL_DOCKER_CTX_DIR}/Nabla")
+make_directory("${NBL_DOCKER_CTX_DIR}/Examples")
 execute_process(
   COMMAND "${CMAKE_COMMAND}" -E copy_if_different "${DXIL_DLL}" "${NBL_DOCKER_CTX_DIR}/Runtimes"
   COMMAND "${CMAKE_COMMAND}" -E copy_if_different "${UCRTBASED_DLL}" "${NBL_DOCKER_CTX_DIR}/Runtimes"
@@ -206,6 +207,9 @@ COPY --link --from=compress ["C:/pack/windows-artifacts.tar.zst", "C:/pack/"]
 COPY --link --from=compress ["C:/pack/nabla-artifacts.tar.zst", "C:/pack/"]
 COPY hlsl.local.properties.cmake C:/Compiler-Explorer/etc/config/hlsl.local.properties
 
+# counting sort
+COPY --link Examples/ C:/Compiler-Explorer/examples/hlsl/
+
 ENV NBL_INSTALL_DIRECTORY=@NBL_DOCKER_CT_NSC_VOLUME_TARGET@ `
 NBL_EXPLICIT_MODULE_LOAD_LOG=ON
 
@@ -262,6 +266,9 @@ endif()
 set(NBL_CE_HEALTHY_CHECK_PY "${NBL_ROOT_PATH}/docker/compiler-explorer/ce_healthy_check.py")
 set(NBL_CE_ENDPOINT_PY "${NBL_ROOT_PATH}/docker/compiler-explorer/endpoint.py")
 set(NBL_NSC_BASIC_HLSL_JPAYLOAD "${CMAKE_CURRENT_SOURCE_DIR}/docker/godbolt/hlsl-basic-compile-payload.json")
+set(NBL_NSC_EXAMPLES_DIR "${CMAKE_CURRENT_SOURCE_DIR}/docker/godbolt/examples")
+
+get_target_property(NBL_CE_COUNTING_SORT_EXAMPLE_PATH countingsort SOURCE_DIR)
 
 # to avoid "too long input" errors we proxy build instructions to CMake script and write it to build directory
 string(CONFIGURE [=[
@@ -306,6 +313,11 @@ execute_process(COMMAND "@CMAKE_COMMAND@" -E copy_directory_if_different
   COMMAND_ERROR_IS_FATAL ANY
 )
 
+message(STATUS "Copying examples")
+execute_process(
+    COMMAND "@CMAKE_COMMAND@" -E copy_directory_if_different "@NBL_NSC_EXAMPLES_DIR@/" "@NBL_DOCKER_CTX_DIR@/Examples/"
+)
+
 message(STATUS "Building NSC Godbolt image")
 string(TIMESTAMP BUILD_TIMESTAMP "%Y-%m-%dT%H:%M:%SZ" UTC)
 execute_process(COMMAND "@DOCKER_EXE@" build --isolation process
@@ -359,4 +371,4 @@ add_custom_target(run-compiler-explorer ALL
 add_dependencies(run-compiler-explorer nsc)
 set_target_properties(run-compiler-explorer PROPERTIES FOLDER "Godbolt")
 
-endif()
+endif()

tools/nsc/docker/godbolt/examples/prefix_sum_shader.comp.hlsl

@@ -0,0 +1,90 @@
+// this shader has to be compiled with:
+// -T cs_6_0 -E main
+
+#include "nbl/builtin/hlsl/sort/counting.hlsl"
+#include "nbl/builtin/hlsl/bda/bda_accessor.hlsl"
+
+#define WorkgroupSize 27
+#define BucketCount 27
+
+struct CountingPushData
+{
+    uint64_t inputKeyAddress;
+    uint64_t inputValueAddress;
+    uint64_t histogramAddress;
+    uint64_t outputKeyAddress;
+    uint64_t outputValueAddress;
+    uint32_t dataElementCount;
+    uint32_t elementsPerWT;
+    uint32_t minimum;
+    uint32_t maximum;
+};
+
+using
+namespace nbl::
+    hlsl;
+
+usingPtr = bda::__ptr<uint32_t>;
+    using PtrAccessor = BdaAccessor < uint32_t >;
+
+    groupshared uint32_t sdata[BucketCount];
+
+    struct SharedAccessor
+    {
+        void get(const uint32_t index, NBL_REF_ARG( uint32_t) value)
+    {
+        value =
+        sdata[ index];
+    }
+
+    void set(
+    const uint32_t index, const uint32_tvalue)
+    {
+        sdata[index] =
+    value;
+}
+
+uint32_t atomicAdd(const uint32_t index, const uint32_t value)
+{
+    return glsl::atomicAdd(sdata[index], value);
+}
+
+void workgroupExecutionAndMemoryBarrier()
+{
+    glsl::barrier();
+}
+};
+
+uint32_t3 glsl::gl_WorkGroupSize()
+{
+    return uint32_t3(WorkgroupSize, 1, 1);
+}
+
+
+[[vk::push_constant]] CountingPushData pushData;
+
+[numthreads(WorkgroupSize, 1, 1)]
+void main(uint32_t3 ID : SV_GroupThreadID, uint32_t3 GroupID : SV_GroupID)
+{
+    sort::CountingParameters < uint32_t > params;
+    params.dataElementCount = pushData.dataElementCount;
+    params.elementsPerWT = pushData.elementsPerWT;
+    params.minimum = pushData.minimum;
+    params.maximum = pushData.maximum;
+
+    using Counter = sort::counting < WorkgroupSize, BucketCount, PtrAccessor, PtrAccessor, PtrAccessor, SharedAccessor, PtrAccessor::type_t >;
+    Counter counter = Counter::create(glsl::gl_WorkGroupID().x);
+
+    const Ptr input_ptr = Ptr::create(pushData.inputKeyAddress);
+    const Ptr histogram_ptr = Ptr::create(pushData.histogramAddress);
+
+    PtrAccessor input_accessor = PtrAccessor::create(input_ptr);
+    PtrAccessor histogram_accessor = PtrAccessor::create(histogram_ptr);
+    SharedAccessor shared_accessor;
+    counter.histogram(
+        input_accessor,
+        histogram_accessor,
+        shared_accessor,
+        params
+    );
+}

tools/nsc/docker/godbolt/examples/scatter_shader.comp.hlsl

@@ -0,0 +1,102 @@
+// this shader has to be compiled with:
+// -T cs_6_0 -E main
+
+#include "nbl/builtin/hlsl/sort/counting.hlsl"
+#include "nbl/builtin/hlsl/bda/bda_accessor.hlsl"
+
+#define WorkgroupSize 27
+#define BucketCount 27
+
+struct CountingPushData
+{
+    uint64_t inputKeyAddress;
+    uint64_t inputValueAddress;
+    uint64_t histogramAddress;
+    uint64_t outputKeyAddress;
+    uint64_t outputValueAddress;
+    uint32_t dataElementCount;
+    uint32_t elementsPerWT;
+    uint32_t minimum;
+    uint32_t maximum;
+};
+
+using
+namespace nbl::
+    hlsl;
+
+usingPtr = bda::__ptr<uint32_t>;
+    using PtrAccessor = BdaAccessor < uint32_t >;
+
+    groupshared uint32_t sdata[BucketCount];
+
+    struct SharedAccessor
+    {
+        void get(const uint32_t index, NBL_REF_ARG( uint32_t) value)
+    {
+        value =
+        sdata[ index];
+    }
+
+    void set(
+    const uint32_t index, const uint32_tvalue)
+    {
+        sdata[index] =
+    value;
+}
+
+uint32_t atomicAdd(const uint32_t index, const uint32_t value)
+{
+    return glsl::atomicAdd(sdata[index], value);
+}
+
+void workgroupExecutionAndMemoryBarrier()
+{
+    glsl::barrier();
+}
+};
+
+uint32_t3 glsl::gl_WorkGroupSize()
+{
+    return uint32_t3(WorkgroupSize, 1, 1);
+}
+
+[[vk::push_constant]] CountingPushData pushData;
+
+using DoublePtrAccessor = DoubleBdaAccessor < uint32_t >;
+
+[numthreads(WorkgroupSize, 1, 1)]
+void main(uint32_t3 ID : SV_GroupThreadID, uint32_t3 GroupID : SV_GroupID)
+{
+    sort::CountingParameters < uint32_t > params;
+    params.dataElementCount = pushData.dataElementCount;
+    params.elementsPerWT = pushData.elementsPerWT;
+    params.minimum = pushData.minimum;
+    params.maximum = pushData.maximum;
+
+    using Counter = sort::counting < WorkgroupSize, BucketCount, DoublePtrAccessor, DoublePtrAccessor, PtrAccessor, SharedAccessor, PtrAccessor::type_t >;
+    Counter counter = Counter::create(glsl::gl_WorkGroupID().x);
+
+    const Ptr input_key_ptr = Ptr::create(pushData.inputKeyAddress);
+    const Ptr input_value_ptr = Ptr::create(pushData.inputValueAddress);
+    const Ptr histogram_ptr = Ptr::create(pushData.histogramAddress);
+    const Ptr output_key_ptr = Ptr::create(pushData.outputKeyAddress);
+    const Ptr output_value_ptr = Ptr::create(pushData.outputValueAddress);
+
+    DoublePtrAccessor key_accessor = DoublePtrAccessor::create(
+        input_key_ptr,
+        output_key_ptr
+    );
+    DoublePtrAccessor value_accessor = DoublePtrAccessor::create(
+        input_value_ptr,
+        output_value_ptr
+    );
+    PtrAccessor histogram_accessor = PtrAccessor::create(histogram_ptr);
+    SharedAccessor shared_accessor;
+    counter.scatter(
+        key_accessor,
+        value_accessor,
+        histogram_accessor,
+        shared_accessor,
+        params
+    );
+}