From 056d59e862a11c9f841a08fbbdfd1a805b098f29 Mon Sep 17 00:00:00 2001
From: assiduous <assiduous@diligentgraphics.com>
Date: Sat, 11 Apr 2020 22:43:32 -0700
Subject: MathLib: improved FastFloat implementation

---
 Common/interface/BasicMath.hpp                    | 23 +++++++++++++++++--
 Tests/DiligentCoreTest/src/Common/MathLibTest.cpp | 28 +++++++++++++++++++++++
 2 files changed, 49 insertions(+), 2 deletions(-)

diff --git a/Common/interface/BasicMath.hpp b/Common/interface/BasicMath.hpp
index 757244e1..25e99240 100644
--- a/Common/interface/BasicMath.hpp
+++ b/Common/interface/BasicMath.hpp
@@ -1942,13 +1942,32 @@ inline Uint32 F4Color_To_RGBA8Unorm(const float4& f4Color)
     return RGBA8U;
 }
 
+template <typename T>
+struct _FastFloatIntermediateType
+{
+};
+
+template <>
+struct _FastFloatIntermediateType<float>
+{
+    // All floats that have fractional part are representable as 32-bit int
+    using Type = Int32;
+};
+
+template <>
+struct _FastFloatIntermediateType<double>
+{
+    // All doubles that have fractional part are representable as 64-bit int
+    using Type = Int64;
+};
 
 // At least on MSVC std::floor is an actual function call into ucrtbase.dll.
-// For floats/doubles that fit into Int64 representable range we can do much better.
+// All floats/doubles that have fractional parts also fit into integer
+// representable range, so we can do much better.
 template <typename T>
 T FastFloor(T x)
 {
-    auto i   = static_cast<Int64>(x);
+    auto i   = static_cast<typename _FastFloatIntermediateType<T>::Type>(x);
     auto flr = static_cast<T>(i);
     //   x         flr    floor(x)  flr <= x
     // +1.0   ->   1.0      1.0       true
diff --git a/Tests/DiligentCoreTest/src/Common/MathLibTest.cpp b/Tests/DiligentCoreTest/src/Common/MathLibTest.cpp
index 4a3da8a2..f7fa48be 100644
--- a/Tests/DiligentCoreTest/src/Common/MathLibTest.cpp
+++ b/Tests/DiligentCoreTest/src/Common/MathLibTest.cpp
@@ -1272,6 +1272,7 @@ TEST(Common_AdvancedMath, TraceLineThroughGrid)
 
 TEST(Common_BasicMath, FastFloor)
 {
+    // float
     EXPECT_EQ(FastFloor(0.f), 0.f);
 
     EXPECT_EQ(FastFloor(-0.0625f), -1.f);
@@ -1283,10 +1284,24 @@ TEST(Common_BasicMath, FastFloor)
     EXPECT_EQ(FastFloor(0.975f), 0.f);
     EXPECT_EQ(FastFloor(1.f), 1.f);
     EXPECT_EQ(FastFloor(1.125f), 1.f);
+
+    // double
+    EXPECT_EQ(FastFloor(0.0), 0.0);
+
+    EXPECT_EQ(FastFloor(-0.03125), -1.0);
+    EXPECT_EQ(FastFloor(-0.96875), -1.0);
+    EXPECT_EQ(FastFloor(-1.0), -1.0);
+    EXPECT_EQ(FastFloor(-1.03125), -2.0);
+
+    EXPECT_EQ(FastFloor(0.03125), 0.0);
+    EXPECT_EQ(FastFloor(0.96875), 0.0);
+    EXPECT_EQ(FastFloor(1.0), 1.0);
+    EXPECT_EQ(FastFloor(1.03125), 1.0);
 }
 
 TEST(Common_BasicMath, FastCeil)
 {
+    // float
     EXPECT_EQ(FastCeil(0.f), 0.f);
 
     EXPECT_EQ(FastCeil(-0.0625f), 0.f);
@@ -1298,6 +1313,19 @@ TEST(Common_BasicMath, FastCeil)
     EXPECT_EQ(FastCeil(0.975f), 1.f);
     EXPECT_EQ(FastCeil(1.f), 1.f);
     EXPECT_EQ(FastCeil(1.125f), 2.f);
+
+    // double
+    EXPECT_EQ(FastCeil(0.0), 0.0);
+
+    EXPECT_EQ(FastCeil(-0.03125), 0.0);
+    EXPECT_EQ(FastCeil(-0.96875), 0.0);
+    EXPECT_EQ(FastCeil(-1.0), -1.0);
+    EXPECT_EQ(FastCeil(-1.03125), -1.0);
+
+    EXPECT_EQ(FastCeil(0.03125), 1.0);
+    EXPECT_EQ(FastCeil(0.96875), 1.0);
+    EXPECT_EQ(FastCeil(1.0), 1.0);
+    EXPECT_EQ(FastCeil(1.03125), 2.0);
 }
 
 } // namespace
-- 
cgit v1.2.3