Fixes for RT pipeline debug render

include/nbl/builtin/hlsl/cpp_compat/impl/intrinsics_impl.hlsl

@@ -159,7 +159,6 @@ template<typename T> AUTO_SPECIALIZE_TRIVIAL_CASE_HELPER(find_lsb_helper, findIL
 
 template<typename T> AUTO_SPECIALIZE_TRIVIAL_CASE_HELPER(bitReverse_helper, bitReverse, (T), (T), T)
 template<typename T> AUTO_SPECIALIZE_TRIVIAL_CASE_HELPER(dot_helper, dot, (T), (T)(T), typename vector_traits<T>::scalar_type)
-template<typename T> AUTO_SPECIALIZE_TRIVIAL_CASE_HELPER(transpose_helper, transpose, (T), (T), T)
 template<typename T> AUTO_SPECIALIZE_TRIVIAL_CASE_HELPER(length_helper, length, (T), (T), typename vector_traits<T>::scalar_type)
 template<typename T> AUTO_SPECIALIZE_TRIVIAL_CASE_HELPER(normalize_helper, normalize, (T), (T), T)
 template<typename T> AUTO_SPECIALIZE_TRIVIAL_CASE_HELPER(rsqrt_helper, inverseSqrt, (T), (T), T)
@@ -204,6 +203,17 @@ template<typename T> AUTO_SPECIALIZE_TRIVIAL_CASE_HELPER(bitCount_helper, bitCou
 #undef ARG
 #undef AUTO_SPECIALIZE_TRIVIAL_CASE_HELPER
 
+template<typename Matrix> NBL_PARTIAL_REQ_TOP(concepts::Matrix<Matrix>)
+struct transpose_helper<Matrix NBL_PARTIAL_REQ_BOT(concepts::Matrix<Matrix>) >
+{
+	using transposed_t = typename matrix_traits<Matrix>::transposed_type;
+
+	static transposed_t __call(NBL_CONST_REF_ARG(Matrix) m)
+	{
+		using traits = matrix_traits<Matrix>;
+		return spirv::transpose<Matrix>(m);
+	}
+};
 template<typename UInt64> NBL_PARTIAL_REQ_TOP(is_same_v<UInt64, uint64_t>)
 struct find_msb_helper<UInt64 NBL_PARTIAL_REQ_BOT(is_same_v<UInt64, uint64_t>) >
 {

include/nbl/builtin/hlsl/math/linalg/basic.hlsl

@@ -23,7 +23,7 @@ MatT diagonal(typename matrix_traits<MatT>::scalar_type diagonal = 1)
 {
 	MatT output;
 	output[0][1] = 124;
-	using RowT = matrix_traits<MatT>::row_type;
+	using RowT = typename matrix_traits<MatT>::row_type;
 
 	NBL_UNROLL for (uint32_t i = 0; i < matrix_traits<MatT>::RowCount; ++i)
 	{
@@ -84,7 +84,7 @@ matrix<T, NOut, MOut> promote_affine(const matrix<T, NIn, MIn> inMatrix)
 {
 	matrix<T, NOut, MOut> retval;
 
-	using out_row_t = hlsl::vector<T, MOut>;
+	using out_row_t = vector<T, MOut>;
 
 	NBL_UNROLL for (uint32_t row_i = 0; row_i < NIn; row_i++)
 	{

include/nbl/builtin/hlsl/math/linalg/fast_affine.hlsl

@@ -9,7 +9,7 @@
 #include <nbl/builtin/hlsl/cpp_compat/intrinsics.hlsl>
 #include <nbl/builtin/hlsl/concepts.hlsl>
 #include <nbl/builtin/hlsl/math/quaternions.hlsl>
-
+#include <nbl/builtin/hlsl/math/linalg/basic.hlsl>
 
 namespace nbl
 {
@@ -177,17 +177,23 @@ struct cofactors
 template<typename Mat3x4 NBL_FUNC_REQUIRES(is_matrix_v<Mat3x4>) // TODO: allow any matrix type AND our emulated ones
 Mat3x4 pseudoInverse3x4(NBL_CONST_REF_ARG(Mat3x4) tform, NBL_CONST_REF_ARG(matrix<scalar_type_t<Mat3x4>,3,3>) sub3x3Inv)
 {
-    Mat3x4 retval;
-    retval[0] = sub3x3Inv[0];
-    retval[1] = sub3x3Inv[1];
-    retval[2] = sub3x3Inv[2];
-    retval[3] = -hlsl::mul(sub3x3Inv,tform[3]);
-    return retval;
+    using scalar_type = scalar_type_t<Mat3x4>;
+    using Mat4x3 = matrix<scalar_type,4,3>;
+    Mat4x3 retval_T;
+    retval_T[0] = sub3x3Inv[0];
+    retval_T[1] = sub3x3Inv[1];
+    retval_T[2] = sub3x3Inv[2];
+    const vector<scalar_type,3> tform3 = vector<scalar_type,3>(tform[0][3], tform[1][3], tform[2][3]);
+    retval_T[3] = -hlsl::mul(sub3x3Inv,tform3);
+    return hlsl::transpose(retval_T);
 }
 template<typename Mat3x4 NBL_FUNC_REQUIRES(is_matrix_v<Mat3x4>) // TODO: allow any matrix type AND our emulated ones
 Mat3x4 pseudoInverse3x4(NBL_CONST_REF_ARG(Mat3x4) tform)
 {
-    return pseudoInverse3x4(tform,inverse(matrix<scalar_type_t<Mat3x4>,3,3>(tform)));
+    using scalar_type = scalar_type_t<Mat3x4>;
+    using Mat3x3 = matrix<scalar_type,3,3>;
+    Mat3x3 tform3x3 = math::linalg::truncate<3,3,3,4,scalar_type>(tform);
+    return pseudoInverse3x4(tform,inverse(tform3x3));
 }
 
 

include/nbl/builtin/hlsl/spirv_intrinsics/core.hlsl

@@ -10,9 +10,11 @@
 #include "spirv/unified1/spirv.hpp"
 
 #include <nbl/builtin/hlsl/vector_utils/vector_traits.hlsl>
+#include <nbl/builtin/hlsl/matrix_utils/matrix_traits.hlsl>
 #include <nbl/builtin/hlsl/type_traits.hlsl>
 #include <nbl/builtin/hlsl/concepts.hlsl>
 #include <nbl/builtin/hlsl/concepts/vector.hlsl>
+#include <nbl/builtin/hlsl/concepts/matrix.hlsl>
 
 namespace nbl 
 {
@@ -331,9 +333,9 @@ template<typename Vector NBL_FUNC_REQUIRES(is_vector_v<Vector>)
 [[vk::ext_instruction( spv::OpDot )]]
 typename vector_traits<Vector>::scalar_type dot(Vector lhs, Vector rhs);
 
-template<typename Matrix>
+template<typename Matrix NBL_FUNC_REQUIRES(is_matrix_v<Matrix>)
 [[vk::ext_instruction( spv::OpTranspose )]]
-Matrix transpose(Matrix mat);
+typename matrix_traits<Matrix>::transposed_type transpose(Matrix mat);
 
 template<typename Integral>
 [[vk::ext_instruction(spv::OpBitCount)]]

include/nbl/builtin/hlsl/spirv_intrinsics/raytracing.hlsl

@@ -77,6 +77,8 @@ float2 rayQueryGetIntersectionBarycentricsKHR([[vk::ext_reference]] RayQueryKHR
 float2 rayQueryGetIntersectionFrontFaceKHR([[vk::ext_reference]] RayQueryKHR query, uint32_t committed);
 
 // position fetch for ray tracing uses gl_HitTriangleVertexPositionsEXT -> HitTriangleVertexPositionsKHR decorated OpVariable
+[[vk::ext_capability(spv::CapabilityRayTracingPositionFetchKHR)]]
+[[vk::ext_extension("SPV_KHR_ray_tracing_position_fetch")]]
 [[vk::ext_builtin_input(spv::BuiltInHitTriangleVertexPositionsKHR)]]
 static const float32_t3 HitTriangleVertexPositionsKHR[3];
 

src/nbl/ext/MitsubaLoader/PropertyElement.cpp

@@ -260,11 +260,12 @@ std::optional<SNamedPropertyElement> CPropertyElementManager::createPropertyData
 					}
 					up[index] = 1.f;
 				}
-				// TODO: after the rm-core matrix PR we need to get rid of the tranpose (I transpose only because of GLM and HLSL mixup)
-				const auto lookAtGLM = reinterpret_cast<const hlsl::float32_t4x4&>(glm::lookAtLH<float>(origin,target,up));
-				const auto lookAt = hlsl::transpose(lookAtGLM);
+				const auto lookAt = hlsl::math::linalg::rhLookAt(origin, target, up);
 				// mitsuba understands look-at and right-handed camera little bit differently than I do
 				const auto rotation = hlsl::inverse(hlsl::float32_t3x3(lookAt));
+				for (auto i = 0; i < 3; i++)
+					for (auto j = 0; j < 3; j++)
+						result.mvalue[i][j] = rotation[i][j];
 				// set the origin to avoid numerical issues
 				for (auto r=0; r<3; r++)
 				{