Fix more bugs with VertexDataLayout driven buffer parsing and generation.

Bump to C++23 standard.

CMakeLists.txt

@@ -1,6 +1,6 @@
 cmake_minimum_required (VERSION 3.16)
 
-set(CMAKE_CXX_STANDARD 20)
+set(CMAKE_CXX_STANDARD 23)
 set(CMAKE_CXX_STANDARD_REQUIRED ON)
 set(CMAKE_CXX_EXTENSIONS OFF)
 

kraken/KREngine-common.h

@@ -120,9 +120,21 @@ using std::multimap;
 using std::stack;
 using std::queue;
 
+#if defined(__x86_64__) || defined(_M_X64)
 
-#if !defined(__i386__) && defined(__arm__)
+  #include <immintrin.h>
+  #define KRAKEN_ARCH_X86_64 1
+
+#elif defined(__aarch64__) || defined(_M_ARM64)
+
+  #include <arm_neon.h>
+  #define KRAKEN_ARCH_ARM64
   #define KRAKEN_USE_ARM_NEON
+
+#else
+
+  static_assert(false, "CPU Architecture not implemented.");
+
 #endif
 
 

kraken/KRPipeline.cpp

@@ -117,13 +117,15 @@ KRPipeline::KRPipeline(KRContext& context, KrDeviceHandle deviceHandle, const KR
           const char* attribBaseName = VertexAttributeNames[(int)attrib.attribute];
           int len = strlen(input_var.name);
           int baseNameLen = strlen(attribBaseName);
-          if (len == baseNameLen) {
+          if (strncmp(attribBaseName, input_var.name, baseNameLen) != 0) {
-            if (strcmp(attribBaseName, input_var.name) == 0) {
+            continue;
-              attribute_locations[i] = input_var.location;
           }
+
+          if (len == baseNameLen) {
+            attribute_locations[i] = input_var.location;
             break;
           }
-          if (len + 1 != baseNameLen) {
+          if (len != baseNameLen + 1) {
             continue;
           }
           char lastChar = input_var.name[len - 1];

kraken/nodes/KRCamera.cpp

@@ -97,7 +97,7 @@ KRCamera::KRCamera(KRScene& scene, std::string name)
   m_debug_text_vbo_layout.offsets[0] = offsetof(DebugTextVertexData, x);
   m_debug_text_vbo_layout.offsets[1] = offsetof(DebugTextVertexData, u);
   m_debug_text_vbo_layout.attributes[0] = { ComponentType::float32, DataType::vec3, Normalization::none, VertexAttribute::position };
-  m_debug_text_vbo_layout.attributes[1] = { ComponentType::float32, DataType::vec2, Normalization::none, VertexAttribute::texcoord };
+  m_debug_text_vbo_layout.attributes[1] = { ComponentType::float16, DataType::vec2, Normalization::none, VertexAttribute::texcoord };
 
   m_debug_text_vbo_data.init(m_pContext->getMeshManager(), &m_debug_text_vertices, nullptr, &m_debug_text_vbo_layout, true, KRMeshManager::KRVBOData::IMMEDIATE
 #if KRENGINE_DEBUG_GPU_LABELS

kraken/nodes/KRCamera.h

@@ -115,8 +115,8 @@ private:
     float x;
     float y;
     float z;
-    float u;
+    short u;
-    float v;
+    short v;
   } DebugTextVertexData;
 
   mimir::Block m_debug_text_vertices;

kraken/resources/mesh/KRMesh.cpp

@@ -490,11 +490,7 @@ void KRMesh::renderSubmesh(VkCommandBuffer& commandBuffer, int iSubmesh, const K
       } else {
         assert(iVertex + cVertexes <= cBufferVertexes);
 
-        if (getIndexCount(0) == 0) {
         vkCmdDraw(commandBuffer, cVertexes, 1, iVertex, 0);
-        } else {
-          vkCmdDrawIndexed(commandBuffer, cVertexes, 1, iVertex, 0, 0);
-        }
         m_pContext->getMeshManager()->log_draw_call(renderPass->getType(), object_name, material_name, cVertexes);
 
         cVertexes = 0;
@@ -510,24 +506,17 @@ void KRMesh::LoadData(const KRMesh::mesh_info& mi, bool calculate_normals, bool
   releaseData();
 
   // TODO, FINDME - These values should be passed as a parameter and set by GUI flags
-  bool use_short_vertexes = false;
+  bool use_f16_vertexes = false;
-  bool use_short_normals = true;
+  bool use_f16_normals = true;
-  bool use_short_tangents = true;
+  bool use_f16_tangents = true;
-  bool use_short_texcoord[8] = { false, false, false, false, false, false, false, false };
+  bool use_f16_texcoord[8] = { false, false, false, false, false, false, false, false };
 
-  if (use_short_vertexes) {
-    for (std::vector<Vector3>::const_iterator itr = mi.vertices.begin(); itr != mi.vertices.end(); itr++) {
-      if (fabsf((*itr).x) > 1.0f || fabsf((*itr).y) > 1.0f || fabsf((*itr).z) > 1.0f) {
-        use_short_vertexes = false;
-      }
-    }
-  }
 
   for (int set = 0; set < 8; set++) {
-    if (use_short_texcoord[set]) {
+    if (use_f16_texcoord[set]) {
       for (std::vector<Vector2>::const_iterator itr = mi.texcoord[set].begin(); itr != mi.texcoord[set].end(); itr++) {
         if (fabsf((*itr).x) > 1.0f || fabsf((*itr).y) > 1.0f) {
-          use_short_texcoord[set] = false;
+          use_f16_texcoord[set] = false;
           break;
         }
       }
@@ -540,36 +529,33 @@ void KRMesh::LoadData(const KRMesh::mesh_info& mi, bool calculate_normals, bool
   if (mi.vertices.size()) {
     attribute->attribute = VertexAttribute::position;
     attribute->type = DataType::vec3;
-    if (use_short_vertexes) {
+    attribute->normalization = Normalization::none;
+    if (use_f16_vertexes) {
       attribute->component = ComponentType::float16;
-      attribute->normalization = Normalization::normalized;
     } else {
       attribute->component = ComponentType::float32;
-      attribute->normalization = Normalization::none;
     }
     attribute++;
   }
   if (mi.normals.size() || calculate_normals) {
     attribute->attribute = VertexAttribute::normal;
     attribute->type = DataType::vec3;
-    if (use_short_normals) {
+    attribute->normalization = Normalization::none;
+    if (use_f16_normals) {
       attribute->component = ComponentType::float16;
-      attribute->normalization = Normalization::normalized;
     } else {
       attribute->component = ComponentType::float32;
-      attribute->normalization = Normalization::none;
     }
     attribute++;
   }
   if (mi.tangents.size() || calculate_tangents) {
     attribute->attribute = VertexAttribute::tangent;
     attribute->type = DataType::vec3;
-    if (use_short_tangents) {
+    attribute->normalization = Normalization::none;
+    if (use_f16_tangents) {
       attribute->component = ComponentType::float16;
-      attribute->normalization = Normalization::normalized;
     } else {
       attribute->component = ComponentType::float32;
-      attribute->normalization = Normalization::none;
     }
     attribute++;
   }
@@ -577,12 +563,11 @@ void KRMesh::LoadData(const KRMesh::mesh_info& mi, bool calculate_normals, bool
     if (mi.texcoord[set].size()) {
       attribute->attribute = VertexAttribute::texcoord;
       attribute->type = DataType::vec2;
-      if (use_short_texcoord[set]) {
+      attribute->normalization = Normalization::none;
+      if (use_f16_texcoord[set]) {
         attribute->component = ComponentType::float16;
-        attribute->normalization = Normalization::normalized;
       } else {
         attribute->component = ComponentType::float32;
-        attribute->normalization = Normalization::none;
       }
       attribute++;
     }
@@ -978,9 +963,13 @@ constexpr void normalizeAttributeComponent(const Normalization norm, float val,
   case Normalization::none:
   case Normalization::scaled:
     {
+      if constexpr (std::is_same_v<T, float>) {
+        *out = val;
+      } else {
         float clampedFloat = std::clamp(val, minFloat, maxFloat);
         *out = static_cast<T>(clampedFloat);
       }
+    }
     break;
   case Normalization::normalized:
     if constexpr (std::is_signed_v<T>) {
@@ -1007,6 +996,111 @@ constexpr void normalizeAttributeComponent(const Normalization norm, float val,
   }
 }
 
+#if defined(KRAKEN_ARCH_X86_64)
+  uint16_t floatToHalf(float val)
+  {
+    __m128 simdFloat = _mm_set_ss(val);
+    __m128i simdHalf = _mm_cvtps_ph(simdFloat, _MM_FROUND_TO_NEAREST_INT);
+    return _mm_cvtsi128_si32(simdHalf) & 0xFFFF;
+  }
+
+  float halfToFloat(uint16_t val)
+  {
+    __m128i simdHalf = _mm_cvtsi32_si128(val);
+    __m128 simdFloat = _mm_cvtph_ps(simdHalf);
+    return _mm_cvtss_f32(simdFloat);
+  }
+
+#elif defined(KRAKEN_ARCH_ARM64)
+  uint16_t floatToHalf(float val)
+  {
+    __fp16 h = static_cast<__fp16>(val);
+    uint16_t bits;
+    std::memcpy(&bits, &h, 2);
+    return bits;
+  }
+
+  // 2. Half to Float
+  float halfToFloat(uint16_t val)
+  {
+    __fp16 h;
+    std::memcpy(&h, &val, 2);
+    return static_cast<float>(h);
+  }
+
+#else
+
+  uint16_t floatToHalf(float val)
+  {
+    uint32_t f32Bits;
+    std::memcpy(&f32Bits, &val, sizeof(float));
+
+    uint32_t sign = (f32Bits >> 16) & 0x8000;
+    int32_t exponent = ((f32Bits >> 23) & 0xFF) - 127;
+    uint32_t mantissa = f32Bits & 0x007FFFFF;
+
+    if (exponent <= -15) {
+      if (exponent < -24) return sign;
+      mantissa |= 0x00800000;
+      return sign | (mantissa >> (-14 - exponent));
+    }
+    if (exponent > 15) return sign | 0x7C00;
+
+    return sign | ((exponent + 15) << 10) | (mantissa >> 13);
+  }
+
+  float halfToFloat(uint16_t val)
+  {
+    uint32_t sign = (val & 0x8000) << 16;
+    uint32_t exponent = (val & 0x7C00) >> 10;
+    uint32_t mantissa = val & 0x03FF;
+    uint32_t f32Bits = 0;
+
+    if (exponent == 0) {
+      if (mantissa != 0) {
+        while ((mantissa & 0x0400) == 0) {
+          mantissa <<= 1;
+          exponent--;
+        }
+        exponent++;
+        mantissa &= 0x03FF;
+        f32Bits = sign | ((exponent + 112) << 23) | (mantissa << 13);
+      } else {
+        f32Bits = sign;
+      }
+    } else if (exponent == 31) {
+      f32Bits = sign | 0x7F800000 | (mantissa << 13);
+    } else {
+      f32Bits = sign | ((exponent + 112) << 23) | (mantissa << 13);
+    }
+
+    float result;
+    std::memcpy(&result, &f32Bits, sizeof(float));
+    return result;
+  }
+
+#endif
+
+// IEEE 754 Half precision specialization
+constexpr void normalizeAttributeComponent_float16(const Normalization norm, float val, uint16_t* out)
+{
+  assert(norm == Normalization::none);
+
+  constexpr float minFloat = -65504.0f;
+  constexpr float maxFloat = 65504.0f;
+
+  float clampedFloat = std::clamp(val, minFloat, maxFloat);
+  *out = floatToHalf(clampedFloat);
+}
+
+// IEEE 754 Half precision specialization
+constexpr void denormalizeAttributeComponent_float16(const Normalization norm, uint16_t val, float* out)
+{
+  assert(norm == Normalization::none);
+
+  *out = halfToFloat(val);
+}
+
 void writeVertexAttributeComponent(const VertexAttributeInfo& attribute, void* address, int componentIndex, float val)
 {
   void* componentAddress = (uint8_t*)address + ComponentSize[(int)attribute.component] * componentIndex;
@@ -1038,7 +1132,7 @@ void writeVertexAttributeComponent(const VertexAttributeInfo& attribute, void* a
     normalizeAttributeComponent(attribute.normalization, val, (__uint64_t*)componentAddress);
     break;
   case ComponentType::float16:
-    normalizeAttributeComponent(attribute.normalization, val, (short*)componentAddress);
+    normalizeAttributeComponent_float16(attribute.normalization, val, (__uint16_t*)componentAddress);
     break;
   case ComponentType::float32:
     normalizeAttributeComponent(attribute.normalization, val, (float*)componentAddress);
@@ -1080,7 +1174,7 @@ void readVertexAttributeComponent(const VertexAttributeInfo& attribute, const vo
     denormalizeAttributeComponent(attribute.normalization, *(__uint64_t*)componentAddress, out);
     break;
   case ComponentType::float16:
-    denormalizeAttributeComponent(attribute.normalization, *(short*)componentAddress, out);
+    denormalizeAttributeComponent_float16(attribute.normalization, *(__uint16_t*)componentAddress, out);
     break;
   case ComponentType::float32:
     denormalizeAttributeComponent(attribute.normalization, *(float*)componentAddress, out);

tests/smoke/hello_cube/CMakeLists.txt

@@ -1,5 +1,5 @@
 cmake_minimum_required (VERSION 3.16)
-set(CMAKE_CXX_STANDARD 11)
+set(CMAKE_CXX_STANDARD 23)
 set(CMAKE_CXX_STANDARD_REQUIRED ON)
 set(CMAKE_CXX_EXTENSIONS OFF)
 

tools/convert/CMakeLists.txt

@@ -1,5 +1,5 @@
 cmake_minimum_required (VERSION 3.16)
-set(CMAKE_CXX_STANDARD 11)
+set(CMAKE_CXX_STANDARD 23)
 set(CMAKE_CXX_STANDARD_REQUIRED ON)
 set(CMAKE_CXX_EXTENSIONS OFF)