Remove SSELegacy.cpp and add validateCPU()

CMakeLists.txt

@@ -27,40 +27,8 @@ set(SOURCES
     src/CAABox.cpp
     src/CEulerAngles.cpp)
 
-# SSELegacy.cpp compiled separately to escape the effects of link-time optimization
-if(NOT MSVC)
-set_source_files_properties(${SOURCES} PROPERTIES COMPILE_FLAGS "-msse4.1 -msse4.2")
-if(CUSTOM_FLAGS)
-    string(REPLACE "-flto=thin" "" CUSTOM_FLAGS ${CMAKE_CXX_FLAGS})
-endif(CUSTOM_FLAGS)
-if (CMAKE_OSX_SYSROOT)
-    set(CUSTOM_FLAGS "${CUSTOM_FLAGS} -isysroot ${CMAKE_OSX_SYSROOT}")
-endif()
-if (CMAKE_OSX_DEPLOYMENT_TARGET AND NOT CMAKE_OSX_DEPLOYMENT_TARGET STREQUAL "")
-    set(CUSTOM_FLAGS "${CUSTOM_FLAGS} -mmacosx-version-min=${CMAKE_OSX_DEPLOYMENT_TARGET} -O3 -msse3 -std=c++14")
-else()
-    set(CUSTOM_FLAGS "${CUSTOM_FLAGS} -O3 -msse3 -std=c++14")
-endif()
-
-set(SSE_LEGACY_OBJ ${CMAKE_CURRENT_BINARY_DIR}/SSELegacy.o)
-separate_arguments(CUSTOM_FLAGS UNIX_COMMAND ${CUSTOM_FLAGS})
-add_custom_command(
-    OUTPUT ${SSE_LEGACY_OBJ}
-    COMMAND ${CMAKE_CXX_COMPILER}
-    ARGS ${CUSTOM_FLAGS} -c ${CMAKE_CURRENT_SOURCE_DIR}/src/SSELegacy.cpp
-                         -o ${SSE_LEGACY_OBJ}
-                         -I ${CMAKE_CURRENT_SOURCE_DIR}/include
-                         -I ${ATHENA_INCLUDE_DIR}
-    MAIN_DEPENDENCY src/SSELegacy.cpp)
-
-else()
-set(SSE_LEGACY_OBJ src/SSELegacy.cpp)
-endif()
-
 add_library(zeus
     ${SOURCES}
-    ${SSE_LEGACY_OBJ}
-
     include/zeus/Math.hpp
     include/zeus/CQuaternion.hpp
     include/zeus/CMatrix3f.hpp

include/zeus/Math.hpp

@@ -48,6 +48,7 @@ struct CPUInfo
  */
 void detectCPU();
 const CPUInfo& cpuFeatures();
+std::pair<bool, const CPUInfo&> validateCPU();
 class CVector3f;
 class CVector2f;
 class CTransform;

src/CMatrix3f.cpp

@@ -33,11 +33,6 @@ CMatrix3f::CMatrix3f(const CQuaternion& quat)
 void CMatrix3f::transpose()
 {
 #if __SSE__
-    if (!cpuFeatures().SSE41)
-    {
-        transposeSSE3();
-        return;
-    }
     __m128 zero = _mm_xor_ps(vec[0].mVec128, vec[0].mVec128);
     __m128 T0 = _mm_unpacklo_ps(vec[0].mVec128, vec[1].mVec128);
     __m128 T2 = _mm_unpacklo_ps(vec[2].mVec128, zero);
@@ -66,8 +61,6 @@ void CMatrix3f::transpose()
 CMatrix3f CMatrix3f::transposed() const
 {
 #if __SSE__
-    if (!cpuFeatures().SSE41)
-        return transposedSSE3();
     __m128 zero = _mm_xor_ps(vec[0].mVec128, vec[0].mVec128);
     __m128 T0 = _mm_unpacklo_ps(vec[0].mVec128, vec[1].mVec128);
     __m128 T2 = _mm_unpacklo_ps(vec[2].mVec128, zero);

src/CMatrix4f.cpp

@@ -9,8 +9,6 @@ CMatrix4f CMatrix4f::transposed() const
 {
     CMatrix4f ret;
 #if __SSE__
-    if (!cpuFeatures().SSE41)
-        return transposedSSE3();
     __m128 T0 = _mm_unpacklo_ps(vec[0].mVec128, vec[1].mVec128);
     __m128 T2 = _mm_unpacklo_ps(vec[2].mVec128, vec[3].mVec128);
     __m128 T1 = _mm_unpackhi_ps(vec[0].mVec128, vec[1].mVec128);

src/CQuaternion.cpp

@@ -48,10 +48,10 @@ CQuaternion CQuaternion::operator*(const CQuaternion& q) const
 
 CNUQuaternion CNUQuaternion::operator*(const CNUQuaternion& q) const
 {
-    return CQuaternion(w * q.w - CVector3f(x, y, z).dot({q.x, q.y, q.z}),
+    return CNUQuaternion(w * q.w - CVector3f(x, y, z).dot({q.x, q.y, q.z}),
-                       y * q.z - z * q.y + w * q.x + x * q.w,
+                         y * q.z - z * q.y + w * q.x + x * q.w,
-                       z * q.x - x * q.z + w * q.y + y * q.w,
+                         z * q.x - x * q.z + w * q.y + y * q.w,
-                       x * q.y - y * q.x + w * q.z + z * q.w);
+                         x * q.y - y * q.x + w * q.z + z * q.w);
 }
 
 CQuaternion CQuaternion::operator/(const CQuaternion& q) const

src/Math.cpp

@@ -12,7 +12,8 @@ namespace zeus
 {
 
 static bool isCPUInit = false;
-static CPUInfo g_cpuFeatures;
+static CPUInfo g_cpuFeatures = {};
+static CPUInfo g_missingFeatures = {};
 
 void getCpuInfo(int level, int regs[4])
 {
@@ -68,6 +69,64 @@ void detectCPU()
 
 const CPUInfo& cpuFeatures() { detectCPU(); return g_cpuFeatures; }
 
+std::pair<bool, const CPUInfo&> validateCPU()
+{
+    detectCPU();
+    bool ret = true;
+
+#if __SSE4A__
+    if (!g_cpuFeatures.SSE4a)
+    {
+        *(bool*) &g_missingFeatures.SSE4a = true;
+        ret = false;
+    }
+#endif
+#if __SSE4_2__
+    if (!g_cpuFeatures.SSE42)
+    {
+        *(bool*) &g_missingFeatures.SSE42 = true;
+        ret = false;
+    }
+#endif
+#if __SSE4_1__
+    if (!g_cpuFeatures.SSE41)
+    {
+        *(bool*) &g_missingFeatures.SSE41 = true;
+        ret = false;
+    }
+#endif
+#if __SSSE3__
+    if (!g_cpuFeatures.SSSE3)
+    {
+        *(bool*) &g_missingFeatures.SSSE3 = true;
+        ret = false;
+    }
+#endif
+#if __SSE3__
+    if (!g_cpuFeatures.SSE3)
+    {
+        *(bool*) &g_missingFeatures.SSE3 = true;
+        ret = false;
+    }
+#endif
+#if __SSE2__
+    if (!g_cpuFeatures.SSE2)
+    {
+        *(bool*) &g_missingFeatures.SSE2 = true;
+        ret = false;
+    }
+#endif
+#if __SSE__
+    if (!g_cpuFeatures.SSE1)
+    {
+        *(bool*) &g_missingFeatures.SSE1 = true;
+        ret = false;
+    }
+#endif
+
+    return {ret, g_missingFeatures};
+}
+
 CTransform lookAt(const CVector3f& pos, const CVector3f& lookPos, const CVector3f& up)
 {
     CVector3f vLook, vRight, vUp;

src/SSELegacy.cpp

@@ -1,99 +0,0 @@
-#include "zeus/CMatrix3f.hpp"
-#include "zeus/CMatrix4f.hpp"
-
-namespace zeus
-{
-
-void CMatrix3f::transposeSSE3()
-{
-#if __SSE__
-    __m128 zero = _mm_xor_ps(vec[0].mVec128, vec[0].mVec128);
-    __m128 T0 = _mm_unpacklo_ps(vec[0].mVec128, vec[1].mVec128);
-    __m128 T2 = _mm_unpacklo_ps(vec[2].mVec128, zero);
-    __m128 T1 = _mm_unpackhi_ps(vec[0].mVec128, vec[1].mVec128);
-    __m128 T3 = _mm_unpackhi_ps(vec[2].mVec128, zero);
-    vec[0].mVec128 = _mm_movelh_ps(T0, T2);
-    vec[1].mVec128 = _mm_movehl_ps(T2, T0);
-    vec[2].mVec128 = _mm_movelh_ps(T1, T3);
-#else
-    float tmp;
-
-    tmp = m[0][1];
-    m[0][1] = m[1][0];
-    m[1][0] = tmp;
-
-    tmp = m[0][2];
-    m[0][2] = m[2][0];
-    m[2][0] = tmp;
-
-    tmp = m[1][2];
-    m[1][2] = m[2][1];
-    m[2][1] = tmp;
-#endif
-}
-
-CMatrix3f CMatrix3f::transposedSSE3() const
-{
-#if __SSE__
-    __m128 zero = _mm_xor_ps(vec[0].mVec128, vec[0].mVec128);
-    __m128 T0 = _mm_unpacklo_ps(vec[0].mVec128, vec[1].mVec128);
-    __m128 T2 = _mm_unpacklo_ps(vec[2].mVec128, zero);
-    __m128 T1 = _mm_unpackhi_ps(vec[0].mVec128, vec[1].mVec128);
-    __m128 T3 = _mm_unpackhi_ps(vec[2].mVec128, zero);
-    return CMatrix3f(_mm_movelh_ps(T0, T2), _mm_movehl_ps(T2, T0), _mm_movelh_ps(T1, T3));
-#else
-    CMatrix3f ret(*this);
-    float tmp;
-
-    tmp = ret.m[0][1];
-    ret.m[0][1] = ret.m[1][0];
-    ret.m[1][0] = tmp;
-
-    tmp = m[0][2];
-    ret.m[0][2] = ret.m[2][0];
-    ret.m[2][0] = tmp;
-
-    tmp = m[1][2];
-    ret.m[1][2] = ret.m[2][1];
-    ret.m[2][1] = tmp;
-
-    return ret;
-#endif
-}
-
-CMatrix4f CMatrix4f::transposedSSE3() const
-{
-    CMatrix4f ret;
-#if __SSE__
-    __m128 T0 = _mm_unpacklo_ps(vec[0].mVec128, vec[1].mVec128);
-    __m128 T2 = _mm_unpacklo_ps(vec[2].mVec128, vec[3].mVec128);
-    __m128 T1 = _mm_unpackhi_ps(vec[0].mVec128, vec[1].mVec128);
-    __m128 T3 = _mm_unpackhi_ps(vec[2].mVec128, vec[3].mVec128);
-    ret.vec[0].mVec128 = _mm_movelh_ps(T0, T2);
-    ret.vec[1].mVec128 = _mm_movehl_ps(T2, T0);
-    ret.vec[2].mVec128 = _mm_movelh_ps(T1, T3);
-    ret.vec[3].mVec128 = _mm_movehl_ps(T3, T1);
-#else
-    ret.m[0][0] = m[0][0];
-    ret.m[1][0] = m[0][1];
-    ret.m[2][0] = m[0][2];
-    ret.m[3][0] = m[0][3];
-
-    ret.m[0][1] = m[1][0];
-    ret.m[1][1] = m[1][1];
-    ret.m[2][1] = m[1][2];
-    ret.m[3][1] = m[1][3];
-
-    ret.m[0][2] = m[2][0];
-    ret.m[1][2] = m[2][1];
-    ret.m[2][2] = m[2][2];
-    ret.m[3][2] = m[2][3];
-
-    ret.m[0][3] = m[3][0];
-    ret.m[1][3] = m[3][1];
-    ret.m[2][3] = m[3][2];
-    ret.m[3][3] = m[3][3];
-#endif
-    return ret;
-}
-}