Merge pull request #848 from Devsh-Graphics-Programming/dxc_spirv_references

Dxc spirv references

Author: Przemog1

3rdparty/dxc/dxc

@@ -113,7 +113,7 @@ struct bound_t
     }
 
 
-    uint operator()(inout Accessor accessor)
+    uint operator()(NBL_REF_ARG(Accessor) accessor)
     {
         if (isNPoT(len))
         {
@@ -132,19 +132,19 @@ struct bound_t
         return it;
     }
 
-    void iteration(inout Accessor accessor)
+    void iteration(NBL_REF_ARG(Accessor) accessor)
     {
         len >>= 1;
         const uint mid = it+len;
         comp_step(accessor,mid);
     }
 
-    void comp_step(inout Accessor accessor, const uint testPoint, const uint rightBegin)
+    void comp_step(NBL_REF_ARG(Accessor) accessor, const uint testPoint, const uint rightBegin)
     {
         if (comp(accessor[testPoint],value))
             it = rightBegin;
     }
-    void comp_step(inout Accessor accessor, const uint testPoint)
+    void comp_step(NBL_REF_ARG(Accessor) accessor, const uint testPoint)
     {
         comp_step(accessor,testPoint,testPoint);
     }
@@ -170,14 +170,14 @@ struct lower_to_upper_comparator_transform_t
 
 
 template<class Accessor, class Comparator>
-uint lower_bound(inout Accessor accessor, const uint begin, const uint end, const typename Accessor::value_type value, const Comparator comp)
+uint lower_bound(NBL_REF_ARG(Accessor) accessor, const uint begin, const uint end, const typename Accessor::value_type value, const Comparator comp)
 {
     impl::bound_t<Accessor,Comparator> implementation = impl::bound_t<Accessor,Comparator>::setup(begin,end,value,comp);
     return implementation(accessor);
 }
 
 template<class Accessor, class Comparator>
-uint upper_bound(inout Accessor accessor, const uint begin, const uint end, const typename Accessor::value_type value, const Comparator comp)
+uint upper_bound(NBL_REF_ARG(Accessor) accessor, const uint begin, const uint end, const typename Accessor::value_type value, const Comparator comp)
 {
     //using TransformedComparator = impl::lower_to_upper_comparator_transform_t<Accessor,Comparator>;
     //TransformedComparator transformedComparator;
@@ -193,7 +193,7 @@ namespace impl
 
 // extra indirection due to https://github.com/microsoft/DirectXShaderCompiler/issues/4771
 template<class Accessor, typename T>
-uint lower_bound(inout Accessor accessor, const uint begin, const uint end, const T value)
+uint lower_bound(NBL_REF_ARG(Accessor) accessor, const uint begin, const uint end, const T value)
 {
     //using Comparator = nbl::hlsl::less<T>;
     //Comparator comp;
@@ -202,7 +202,7 @@ uint lower_bound(inout Accessor accessor, const uint begin, const uint end, cons
     return nbl::hlsl::lower_bound<Accessor, nbl::hlsl::less<T> >(accessor,begin,end,value,comp);
 }
 template<class Accessor, typename T>
-uint upper_bound(inout Accessor accessor, const uint begin, const uint end, const T value)
+uint upper_bound(NBL_REF_ARG(Accessor) accessor, const uint begin, const uint end, const T value)
 {
     //using Comparator = nbl::hlsl::less<T>;
     //Comparator comp;
@@ -214,12 +214,12 @@ uint upper_bound(inout Accessor accessor, const uint begin, const uint end, cons
 }
 
 template<class Accessor>
-uint lower_bound(inout Accessor accessor, const uint begin, const uint end, const typename Accessor::value_type value)
+uint lower_bound(NBL_REF_ARG(Accessor) accessor, const uint begin, const uint end, const typename Accessor::value_type value)
 {
     return impl::lower_bound<Accessor,typename Accessor::value_type>(accessor,begin,end,value);
 }
 template<class Accessor>
-uint upper_bound(inout Accessor accessor, const uint begin, const uint end, const typename Accessor::value_type value)
+uint upper_bound(NBL_REF_ARG(Accessor) accessor, const uint begin, const uint end, const typename Accessor::value_type value)
 {
     return impl::upper_bound<Accessor,typename Accessor::value_type>(accessor,begin,end,value);
 }

include/nbl/builtin/hlsl/algorithm.hlsl

@@ -37,10 +37,9 @@ enable_if_t<sizeof(T)==sizeof(U)&&(is_scalar_v<T>||is_vector_v<T>)==(is_scalar_v
 {
     return spirv::bitcast<T,U>(val);
 }
-// unfortunately its impossible to deduce Storage Class right now,
-// also this function will only work as long as `inout` behaves as `__restrict &` in DXC
+// unfortunately its impossible to deduce Storage Class right now
 template<class T, class U, uint32_t StorageClass>
-enable_if_t<sizeof(T)==sizeof(U),T> bit_cast(inout U val)
+enable_if_t<sizeof(T)==sizeof(U),T> bit_cast(NBL_REF_ARG(U) val)
 {
     using ptr_u_t = spirv::pointer_t<U,StorageClass>;
     // get the address of U

include/nbl/builtin/hlsl/bit.hlsl

@@ -13,7 +13,7 @@ namespace workgroup
 namespace fft
 {
 // The SharedMemoryAccessor MUST provide the following methods:
-//      * void get(uint32_t index, inout uint32_t value);  
+//      * void get(uint32_t index, NBL_REF_ARG(uint32_t) value);  
 //      * void set(uint32_t index, in uint32_t value); 
 //      * void workgroupExecutionAndMemoryBarrier();
 
@@ -39,7 +39,7 @@ NBL_CONCEPT_END(
 
 
 // The Accessor (for a small FFT) MUST provide the following methods:
-//     * void get(uint32_t index, inout complex_t<Scalar> value);
+//     * void get(uint32_t index, NBL_REF_ARG(complex_t<Scalar>) value);
 //     * void set(uint32_t index, in complex_t<Scalar> value);
 
 #define NBL_CONCEPT_NAME FFTAccessor

include/nbl/builtin/hlsl/concepts/accessors/fft.hlsl

@@ -95,7 +95,7 @@ struct add_pointer
 }
 }
 
-#define NBL_REF_ARG(...) inout __VA_ARGS__
+#define NBL_REF_ARG(...) [[vk::ext_reference]] inout __VA_ARGS__
 #define NBL_CONST_REF_ARG(...) const in __VA_ARGS__
 
 #endif

include/nbl/builtin/hlsl/cpp_compat/basic.h

@@ -28,7 +28,7 @@ struct result_t
 // Elements with value 0 do not get appended
 // Note: If NBL_GLSL_EXT_shader_atomic_int64 is not present, then the call to these functions needs to be subgroup uniform
 template<class AtomicCounterAccessor>
-result_t non_negative(inout AtomicCounterAccessor accessor, in uint value)
+result_t non_negative(NBL_REF_ARG(AtomicCounterAccessor) accessor, in uint value)
 {
   const bool willAppend = bool(value);
 
@@ -70,7 +70,7 @@ result_t non_negative(inout AtomicCounterAccessor accessor, in uint value)
 
 // optimized version which tries to omit the atomicAdd and locks if it can, in return it may return garbage/invalid value when invocation's `value==0`
 template<class AtomicCounterAccessor>
-result_t positive(inout AtomicCounterAccessor accessor, in uint value)
+result_t positive(NBL_REF_ARG(AtomicCounterAccessor) accessor, in uint value)
 {
   const bool willAppend = bool(value);
 #ifdef NBL_GLSL_EXT_shader_atomic_int64

include/nbl/builtin/hlsl/scanning_append.hlsl

@@ -294,7 +294,7 @@ struct FFT;
 //        If `ConstevalParameters::ElementsPerInvocationLog2 == 1`, the arrays it accesses with `get` and `set` can optionally be different, 
 //        if you don't want the FFT to be done in-place. Otherwise, you MUST make it in-place.
 //        The Accessor MUST provide the following methods:
-//            * void get(uint32_t index, inout complex_t<Scalar> value);
+//            * void get(uint32_t index, NBL_REF_ARG(complex_t<Scalar>) value);
 //            * void set(uint32_t index, in complex_t<Scalar> value);
 //            * void memoryBarrier();
 //        For it to work correctly, this memory barrier must use `AcquireRelease` semantics, with the proper flags set for the memory type.
@@ -303,7 +303,7 @@ struct FFT;
 
 //      - SharedMemoryAccessor accesses a workgroup-shared memory array of size `WorkgroupSize` elements of type complex_t<Scalar>.
 //        The SharedMemoryAccessor MUST provide the following methods:
-//             * void get(uint32_t index, inout uint32_t value);  
+//             * void get(uint32_t index, NBL_REF_ARG(uint32_t) value);  
 //             * void set(uint32_t index, in uint32_t value); 
 //             * void workgroupExecutionAndMemoryBarrier();
 

include/nbl/builtin/hlsl/workgroup/fft.hlsl

@@ -24,7 +24,7 @@ using namespace nbl;
 using namespace nbl::asset;
 using Microsoft::WRL::ComPtr;
 
-static constexpr const wchar_t* SHADER_MODEL_PROFILE = L"XX_6_7";
+static constexpr const wchar_t* SHADER_MODEL_PROFILE = L"XX_6_8";
 static const wchar_t* ShaderStageToString(asset::IShader::E_SHADER_STAGE stage) {
     switch (stage)
     {
@@ -128,14 +128,15 @@ static void try_upgrade_hlsl_version(std::vector<std::wstring>& arguments, syste
 
 static void try_upgrade_shader_stage(std::vector<std::wstring>& arguments, asset::IShader::E_SHADER_STAGE shaderStageOverrideFromPragma, system::logger_opt_ptr& logger) {
 
-    constexpr int MajorReqVersion = 6, MinorReqVersion = 7;
+    constexpr int MajorReqVersion = 6, MinorReqVersion = 8;
     auto overrideStageStr = ShaderStageToString(shaderStageOverrideFromPragma);
     if (shaderStageOverrideFromPragma != IShader::E_SHADER_STAGE::ESS_UNKNOWN && !overrideStageStr)
     {
         logger.log("Invalid shader stage with int value '%i'.\nThis value does not have a known string representation.",
             system::ILogger::ELL_ERROR, shaderStageOverrideFromPragma);
         return;
     }
+    // still unknown, then we take value from commandline arguments (precedence: pragma > DXC specific command line > compile option)
     bool setDefaultValue = true;
     auto foundShaderStageArgument = std::find(arguments.begin(), arguments.end(), L"-T");
     if (foundShaderStageArgument != arguments.end() && foundShaderStageArgument +1 != arguments.end()) {
@@ -202,6 +203,9 @@ static void try_upgrade_shader_stage(std::vector<std::wstring>& arguments, asset
     }
     if (setDefaultValue) 
     { 
+        // if stage is still not known, lets go with library
+        if (shaderStageOverrideFromPragma==IShader::E_SHADER_STAGE::ESS_UNKNOWN)
+            overrideStageStr = ShaderStageToString(hlsl::ShaderStage::ESS_ALL_OR_LIBRARY);
         // in case of no -T
         // push back default values for -T argument
         // can be safely pushed to the back of argument list as output files should be evicted from args before passing to this func
@@ -379,6 +383,7 @@ std::string CHLSLCompiler::preprocessShader(std::string&& code, IShader::E_SHADE
     if (context.get_hooks().m_dxc_compile_flags_override.size() != 0)
         dxc_compile_flags_override = context.get_hooks().m_dxc_compile_flags_override;
 
+    // pragma overrides what we passed in
     if(context.get_hooks().m_pragmaStage != IShader::E_SHADER_STAGE::ESS_UNKNOWN)
         stage = context.get_hooks().m_pragmaStage;
 

src/nbl/asset/utils/CHLSLCompiler.cpp

@@ -45,6 +45,7 @@ core::smart_refctd_ptr<ICPUShader> nbl::asset::IShaderCompiler::compileToSPIRV(c
 
     auto retVal = compileToSPIRV_impl(code, options, options.writeCache ? &entry.dependencies : nullptr);
     // compute the SPIR-V shader content hash
+    if (retVal)
     {
         auto backingBuffer = retVal->getContent();
         const_cast<ICPUBuffer*>(backingBuffer)->setContentHash(backingBuffer->computeContentHash());

src/nbl/asset/utils/IShaderCompiler.cpp

@@ -359,5 +359,8 @@ LIST_BUILTIN_RESOURCE(NBL_RESOURCES_TO_EMBED "hlsl/concepts/accessors/fft.hlsl")
 LIST_BUILTIN_RESOURCE(NBL_RESOURCES_TO_EMBED "hlsl/tgmath.hlsl")
 LIST_BUILTIN_RESOURCE(NBL_RESOURCES_TO_EMBED "hlsl/tgmath/impl.hlsl")
 LIST_BUILTIN_RESOURCE(NBL_RESOURCES_TO_EMBED "hlsl/tgmath/output_structs.hlsl")
+#blur
+LIST_BUILTIN_RESOURCE(NBL_RESOURCES_TO_EMBED "hlsl/prefix_sum_blur/blur.hlsl")
+LIST_BUILTIN_RESOURCE(NBL_RESOURCES_TO_EMBED "hlsl/prefix_sum_blur/box_sampler.hlsl")
 
 ADD_CUSTOM_BUILTIN_RESOURCES(nblBuiltinResourceData NBL_RESOURCES_TO_EMBED "${NBL_ROOT_PATH}/include" "nbl/builtin" "nbl::builtin" "${NBL_ROOT_PATH_BINARY}/include" "${NBL_ROOT_PATH_BINARY}/src" "STATIC" "INTERNAL")