Update version to 1.1.4

* Handle zero-vector normalization.

When normalizing a vectors, we have to check whether vector length is
not zero, to avoid dividing by zero when normalizing zero-vectors.

* Test for normalization of zero vectors

.travis.yml

@@ -7,4 +7,6 @@ install:
   - make
 script:
   - ./hmm_test_c
+  - ./hmm_test_c_no_sse
   - ./hmm_test_cpp
+  - ./hmm_test_cpp_no_sse

HandmadeMath.h

@@ -1,5 +1,5 @@
 /*
-  HandmadeMath.h v1.1.2
+  HandmadeMath.h v1.1.4
   
   This is a single header file with a bunch of useful functions for
   basic game math operations.
@@ -175,6 +175,8 @@
           (*) Fixed invalid HMMDEF's in the function definitions
      1.1.3
           (*) Fixed compile error in C mode
+     1.1.4
+          (*) Fixed divide-by-zero errors when normalizing zero vectors.
           
   LICENSE
   
@@ -200,7 +202,27 @@
    Insofaras (@insofaras)
 */
 
-#ifndef HANDMADE_NO_SSE
+
+/* let's figure out if SSE is really available (unless disabled anyway)
+   (it isn't on non-x86/x86_64 platforms or even x86 without explicit SSE support)
+   => only use "#ifdef HANDMADE_MATH__USE_SSE" to check for SSE support below this block! */
+#ifndef HANDMADE_MATH_NO_SSE
+
+# ifdef _MSC_VER
+   /* MSVC supports SSE in amd64 mode or _M_IX86_FP >= 1 (2 means SSE2) */
+#  if defined(_M_AMD64) || ( defined(_M_IX86_FP) && _M_IX86_FP >= 1 )
+#   define HANDMADE_MATH__USE_SSE 1
+#  endif
+# else /* not MSVC, probably GCC, clang, icc or something that doesn't support SSE anyway */
+#  ifdef __SSE__ /* they #define __SSE__ if it's supported */
+#   define HANDMADE_MATH__USE_SSE 1
+#  endif /*  __SSE__ */
+# endif /* not _MSC_VER */
+
+#endif /* #ifndef HANDMADE_MATH_NO_SSE */
+
+
+#ifdef HANDMADE_MATH__USE_SSE
 #include <xmmintrin.h>
 #endif
 
@@ -757,12 +779,12 @@ HMM_SquareRootF(float Value)
 {
     float Result = 0.0f;
 
-#ifdef HANDMADE_MATH_NO_SSE
-    Result = sqrtf(Value);
-#else        
+#ifdef HANDMADE_MATH__USE_SSE
     __m128 In = _mm_set_ss(Value);
     __m128 Out = _mm_sqrt_ss(In);
     Result = _mm_cvtss_f32(Out);
+#else
+    Result = sqrtf(Value);
 #endif 
 
     return(Result);
@@ -773,12 +795,12 @@ HMM_RSquareRootF(float Value)
 {
     float Result = 0.0f;
 
-#ifdef HANDMADE_MATH_NO_SSE
-    Result = 1.0f/HMM_SqrtF(Value);    
-#else        
+#ifdef HANDMADE_MATH__USE_SSE
     __m128 In = _mm_set_ss(Value);
     __m128 Out = _mm_rsqrt_ss(In);
     Result = _mm_cvtss_f32(Out);
+#else
+    Result = 1.0f/HMM_SquareRootF(Value);
 #endif
 
     return(Result);
@@ -903,8 +925,12 @@ HMM_NormalizeVec2(hmm_vec2 A)
 
     float VectorLength = HMM_LengthVec2(A);
     
-    Result.X = A.X * (1.0f / VectorLength);
-    Result.Y = A.Y * (1.0f / VectorLength);
+    /* NOTE(kiljacken): We need a zero check to not divide-by-zero */
+    if (VectorLength != 0.0f)
+    {
+        Result.X = A.X * (1.0f / VectorLength);
+        Result.Y = A.Y * (1.0f / VectorLength);
+    }
     
     return (Result);
 }
@@ -916,9 +942,13 @@ HMM_NormalizeVec3(hmm_vec3 A)
 
     float VectorLength = HMM_LengthVec3(A);
     
-    Result.X = A.X * (1.0f / VectorLength);
-    Result.Y = A.Y * (1.0f / VectorLength);
-    Result.Z = A.Z * (1.0f / VectorLength);
+    /* NOTE(kiljacken): We need a zero check to not divide-by-zero */
+    if (VectorLength != 0.0f)
+    {
+        Result.X = A.X * (1.0f / VectorLength);
+        Result.Y = A.Y * (1.0f / VectorLength);
+        Result.Z = A.Z * (1.0f / VectorLength);
+    }
     
     return (Result);
 }
@@ -930,10 +960,14 @@ HMM_NormalizeVec4(hmm_vec4 A)
 
     float VectorLength = HMM_LengthVec4(A);
     
-    Result.X = A.X * (1.0f / VectorLength);
-    Result.Y = A.Y * (1.0f / VectorLength);
-    Result.Z = A.Z * (1.0f / VectorLength);
-    Result.W = A.W * (1.0f / VectorLength);
+    /* NOTE(kiljacken): We need a zero check to not divide-by-zero */
+    if (VectorLength != 0.0f)
+    {
+        Result.X = A.X * (1.0f / VectorLength);
+        Result.Y = A.Y * (1.0f / VectorLength);
+        Result.Z = A.Z * (1.0f / VectorLength);
+        Result.W = A.W * (1.0f / VectorLength);
+    }
     
     return (Result);
 }

README.md

@@ -12,6 +12,7 @@ _This library is free and will stay free, but if you would like to support devel
 
 Version         | Changes        | 
 ----------------|----------------|
+**1.1.4** | Fixed divide-by-zero errors when normalizing zero vectors.
 **1.1.3** | Fixed compile error in C mode
 **1.1.2** | Fixed invalid HMMDEF's in the function definitions
 **1.1.1** | Resolved compiler warnings on gcc and g++

test/HandmadeMath_NoSSE.c

@@ -0,0 +1,5 @@
+
+#define HANDMADE_MATH_IMPLEMENTATION
+#define HANDMADE_MATH_NO_SSE
+#define HANDMADE_MATH_NO_INLINE
+#include "../HandmadeMath.h"

test/HandmadeMath_NoSSE.cpp

@@ -0,0 +1,6 @@
+
+#define HANDMADE_MATH_IMPLEMENTATION
+#define HANDMADE_MATH_CPP_MODE
+#define HANDMADE_MATH_NO_SSE
+#define HANDMADE_MATH_NO_INLINE
+#include "../HandmadeMath.h"

test/Makefile

@@ -2,16 +2,23 @@ ROOT_DIR = ..
 
 CXXFLAGS += -g -Wall -Wextra -pthread -Wno-missing-braces -Wno-missing-field-initializers
 
-all: c cpp
+all: c c_no_sse cpp cpp_no_sse
 
 clean:
-	rm -f hmm_test_c hmm_test_cpp *.o
+	rm -f hmm_test_c hmm_test_cpp hmm_test_c_no_sse hmm_test_cpp_no_sse *.o
 
 c: $(ROOT_DIR)/test/HandmadeMath.c test_impl
 	$(CC) $(CPPFLAGS) $(CXXFLAGS) -std=c99 -c $(ROOT_DIR)/test/HandmadeMath.c $(ROOT_DIR)/test/hmm_test.c -lm
 	$(CC) -ohmm_test_c HandmadeMath.o hmm_test.o -lm
 
+c_no_sse: $(ROOT_DIR)/test/HandmadeMath_NoSSE.c test_impl
+	$(CC) $(CPPFLAGS) $(CXXFLAGS) -std=c99 -c $(ROOT_DIR)/test/HandmadeMath_NoSSE.c $(ROOT_DIR)/test/hmm_test.c -lm
+	$(CC) -ohmm_test_c_no_sse HandmadeMath_NoSSE.o hmm_test.o -lm
+
 cpp: $(ROOT_DIR)/test/HandmadeMath.cpp test_impl
 	$(CXX) $(CPPFLAGS) $(CXXFLAGS) -ohmm_test_cpp $(ROOT_DIR)/test/HandmadeMath.cpp $(ROOT_DIR)/test/hmm_test.cpp
 
+cpp_no_sse: $(ROOT_DIR)/test/HandmadeMath_NoSSE.cpp test_impl
+	$(CXX) $(CPPFLAGS) $(CXXFLAGS) -ohmm_test_cpp_no_sse $(ROOT_DIR)/test/HandmadeMath_NoSSE.cpp $(ROOT_DIR)/test/hmm_test.cpp
+
 test_impl: $(ROOT_DIR)/test/hmm_test.cpp $(ROOT_DIR)/test/hmm_test.c

test/hmm_test.c

@@ -70,7 +70,7 @@ int run_tests()
 
         TEST_BEGIN(RSquareRootF)
         {
-            EXPECT_FLOAT_EQ(HMM_RSquareRootF(10.0f), 0.31616211f);
+            EXPECT_NEAR(HMM_RSquareRootF(10.0f), 0.31616211f, 0.0001f);
         }
         TEST_END()
 
@@ -310,6 +310,54 @@ int run_tests()
         }
         TEST_END()
 
+        TEST_BEGIN(NormalizeZero)
+        {
+            hmm_vec2 v2 = HMM_Vec2(0.0f, 0.0f);
+            hmm_vec3 v3 = HMM_Vec3(0.0f, 0.0f, 0.0f);
+            hmm_vec4 v4 = HMM_Vec4(0.0f, 0.0f, 0.0f, 0.0f);
+
+            {
+                hmm_vec2 result = HMM_NormalizeVec2(v2);
+                EXPECT_FLOAT_EQ(result.X, 0.0f);
+                EXPECT_FLOAT_EQ(result.Y, 0.0f);
+            }
+            {
+                hmm_vec3 result = HMM_NormalizeVec3(v3);
+                EXPECT_FLOAT_EQ(result.X, 0.0f);
+                EXPECT_FLOAT_EQ(result.Y, 0.0f);
+                EXPECT_FLOAT_EQ(result.Z, 0.0f);
+            }
+            {
+                hmm_vec4 result = HMM_NormalizeVec4(v4);
+                EXPECT_FLOAT_EQ(result.X, 0.0f);
+                EXPECT_FLOAT_EQ(result.Y, 0.0f);
+                EXPECT_FLOAT_EQ(result.Z, 0.0f);
+                EXPECT_FLOAT_EQ(result.W, 0.0f);
+            }
+
+#ifdef HANDMADE_MATH_CPP_MODE
+            {
+                hmm_vec2 result = HMM_Normalize(v2);
+                EXPECT_FLOAT_EQ(result.X, 0.0f);
+                EXPECT_FLOAT_EQ(result.Y, 0.0f);
+            }
+            {
+                hmm_vec3 result = HMM_Normalize(v3);
+                EXPECT_FLOAT_EQ(result.X, 0.0f);
+                EXPECT_FLOAT_EQ(result.Y, 0.0f);
+                EXPECT_FLOAT_EQ(result.Z, 0.0f);
+            }
+            {
+                hmm_vec4 result = HMM_Normalize(v4);
+                EXPECT_FLOAT_EQ(result.X, 0.0f);
+                EXPECT_FLOAT_EQ(result.Y, 0.0f);
+                EXPECT_FLOAT_EQ(result.Z, 0.0f);
+                EXPECT_FLOAT_EQ(result.W, 0.0f);
+            }
+#endif
+        }
+        TEST_END()
+
         TEST_BEGIN(Cross)
         {
             hmm_vec3 v1 = HMM_Vec3(1.0f, 2.0f, 3.0f);