aboutsummaryrefslogtreecommitdiff
path: root/3rdparty/glm/source/test/ext/ext_scalar_integer.cpp
diff options
context:
space:
mode:
Diffstat (limited to '3rdparty/glm/source/test/ext/ext_scalar_integer.cpp')
-rw-r--r--3rdparty/glm/source/test/ext/ext_scalar_integer.cpp686
1 files changed, 686 insertions, 0 deletions
diff --git a/3rdparty/glm/source/test/ext/ext_scalar_integer.cpp b/3rdparty/glm/source/test/ext/ext_scalar_integer.cpp
new file mode 100644
index 0000000..f169e8a
--- /dev/null
+++ b/3rdparty/glm/source/test/ext/ext_scalar_integer.cpp
@@ -0,0 +1,686 @@
+#include <glm/ext/scalar_integer.hpp>
+#include <glm/ext/scalar_int_sized.hpp>
+#include <glm/ext/scalar_uint_sized.hpp>
+#include <vector>
+#include <ctime>
+#include <cstdio>
+
+#if GLM_LANG & GLM_LANG_CXX11_FLAG
+#include <chrono>
+
+namespace isPowerOfTwo
+{
+ template<typename genType>
+ struct type
+ {
+ genType Value;
+ bool Return;
+ };
+
+ int test_int16()
+ {
+ type<glm::int16> const Data[] =
+ {
+ {0x0001, true},
+ {0x0002, true},
+ {0x0004, true},
+ {0x0080, true},
+ {0x0000, true},
+ {0x0003, false}
+ };
+
+ int Error = 0;
+
+ for(std::size_t i = 0, n = sizeof(Data) / sizeof(type<glm::int16>); i < n; ++i)
+ {
+ bool Result = glm::isPowerOfTwo(Data[i].Value);
+ Error += Data[i].Return == Result ? 0 : 1;
+ }
+
+ return Error;
+ }
+
+ int test_uint16()
+ {
+ type<glm::uint16> const Data[] =
+ {
+ {0x0001, true},
+ {0x0002, true},
+ {0x0004, true},
+ {0x0000, true},
+ {0x0000, true},
+ {0x0003, false}
+ };
+
+ int Error = 0;
+
+ for(std::size_t i = 0, n = sizeof(Data) / sizeof(type<glm::uint16>); i < n; ++i)
+ {
+ bool Result = glm::isPowerOfTwo(Data[i].Value);
+ Error += Data[i].Return == Result ? 0 : 1;
+ }
+
+ return Error;
+ }
+
+ int test_int32()
+ {
+ type<int> const Data[] =
+ {
+ {0x00000001, true},
+ {0x00000002, true},
+ {0x00000004, true},
+ {0x0000000f, false},
+ {0x00000000, true},
+ {0x00000003, false}
+ };
+
+ int Error = 0;
+
+ for(std::size_t i = 0, n = sizeof(Data) / sizeof(type<int>); i < n; ++i)
+ {
+ bool Result = glm::isPowerOfTwo(Data[i].Value);
+ Error += Data[i].Return == Result ? 0 : 1;
+ }
+
+ return Error;
+ }
+
+ int test_uint32()
+ {
+ type<glm::uint> const Data[] =
+ {
+ {0x00000001, true},
+ {0x00000002, true},
+ {0x00000004, true},
+ {0x80000000, true},
+ {0x00000000, true},
+ {0x00000003, false}
+ };
+
+ int Error = 0;
+
+ for(std::size_t i = 0, n = sizeof(Data) / sizeof(type<glm::uint>); i < n; ++i)
+ {
+ bool Result = glm::isPowerOfTwo(Data[i].Value);
+ Error += Data[i].Return == Result ? 0 : 1;
+ }
+
+ return Error;
+ }
+
+ int test()
+ {
+ int Error = 0;
+
+ Error += test_int16();
+ Error += test_uint16();
+ Error += test_int32();
+ Error += test_uint32();
+
+ return Error;
+ }
+}//isPowerOfTwo
+
+namespace nextPowerOfTwo_advanced
+{
+ template<typename genIUType>
+ GLM_FUNC_QUALIFIER genIUType highestBitValue(genIUType Value)
+ {
+ genIUType tmp = Value;
+ genIUType result = genIUType(0);
+ while(tmp)
+ {
+ result = (tmp & (~tmp + 1)); // grab lowest bit
+ tmp &= ~result; // clear lowest bit
+ }
+ return result;
+ }
+
+ template<typename genType>
+ GLM_FUNC_QUALIFIER genType nextPowerOfTwo_loop(genType value)
+ {
+ return glm::isPowerOfTwo(value) ? value : highestBitValue(value) << 1;
+ }
+
+ template<typename genType>
+ struct type
+ {
+ genType Value;
+ genType Return;
+ };
+
+ int test_int32()
+ {
+ type<glm::int32> const Data[] =
+ {
+ {0x0000ffff, 0x00010000},
+ {-3, -4},
+ {-8, -8},
+ {0x00000001, 0x00000001},
+ {0x00000002, 0x00000002},
+ {0x00000004, 0x00000004},
+ {0x00000007, 0x00000008},
+ {0x0000fff0, 0x00010000},
+ {0x0000f000, 0x00010000},
+ {0x08000000, 0x08000000},
+ {0x00000000, 0x00000000},
+ {0x00000003, 0x00000004}
+ };
+
+ int Error(0);
+
+ for(std::size_t i = 0, n = sizeof(Data) / sizeof(type<glm::int32>); i < n; ++i)
+ {
+ glm::int32 Result = glm::nextPowerOfTwo(Data[i].Value);
+ Error += Data[i].Return == Result ? 0 : 1;
+ }
+
+ return Error;
+ }
+
+ int test_uint32()
+ {
+ type<glm::uint32> const Data[] =
+ {
+ {0x00000001, 0x00000001},
+ {0x00000002, 0x00000002},
+ {0x00000004, 0x00000004},
+ {0x00000007, 0x00000008},
+ {0x0000ffff, 0x00010000},
+ {0x0000fff0, 0x00010000},
+ {0x0000f000, 0x00010000},
+ {0x80000000, 0x80000000},
+ {0x00000000, 0x00000000},
+ {0x00000003, 0x00000004}
+ };
+
+ int Error(0);
+
+ for(std::size_t i = 0, n = sizeof(Data) / sizeof(type<glm::uint32>); i < n; ++i)
+ {
+ glm::uint32 Result = glm::nextPowerOfTwo(Data[i].Value);
+ Error += Data[i].Return == Result ? 0 : 1;
+ }
+
+ return Error;
+ }
+
+ int perf()
+ {
+ int Error(0);
+
+ std::vector<glm::uint> v;
+ v.resize(100000000);
+
+ std::clock_t Timestramp0 = std::clock();
+
+ for(glm::uint32 i = 0, n = static_cast<glm::uint>(v.size()); i < n; ++i)
+ v[i] = nextPowerOfTwo_loop(i);
+
+ std::clock_t Timestramp1 = std::clock();
+
+ for(glm::uint32 i = 0, n = static_cast<glm::uint>(v.size()); i < n; ++i)
+ v[i] = glm::nextPowerOfTwo(i);
+
+ std::clock_t Timestramp2 = std::clock();
+
+ std::printf("nextPowerOfTwo_loop: %d clocks\n", static_cast<int>(Timestramp1 - Timestramp0));
+ std::printf("glm::nextPowerOfTwo: %d clocks\n", static_cast<int>(Timestramp2 - Timestramp1));
+
+ return Error;
+ }
+
+ int test()
+ {
+ int Error(0);
+
+ Error += test_int32();
+ Error += test_uint32();
+
+ return Error;
+ }
+}//namespace nextPowerOfTwo_advanced
+
+namespace prevPowerOfTwo
+{
+ template <typename T>
+ int run()
+ {
+ int Error = 0;
+
+ T const A = glm::prevPowerOfTwo(static_cast<T>(7));
+ Error += A == static_cast<T>(4) ? 0 : 1;
+
+ T const B = glm::prevPowerOfTwo(static_cast<T>(15));
+ Error += B == static_cast<T>(8) ? 0 : 1;
+
+ T const C = glm::prevPowerOfTwo(static_cast<T>(31));
+ Error += C == static_cast<T>(16) ? 0 : 1;
+
+ T const D = glm::prevPowerOfTwo(static_cast<T>(32));
+ Error += D == static_cast<T>(32) ? 0 : 1;
+
+ return Error;
+ }
+
+ int test()
+ {
+ int Error = 0;
+
+ Error += run<glm::int8>();
+ Error += run<glm::int16>();
+ Error += run<glm::int32>();
+ Error += run<glm::int64>();
+
+ Error += run<glm::uint8>();
+ Error += run<glm::uint16>();
+ Error += run<glm::uint32>();
+ Error += run<glm::uint64>();
+
+ return Error;
+ }
+}//namespace prevPowerOfTwo
+
+namespace nextPowerOfTwo
+{
+ template <typename T>
+ int run()
+ {
+ int Error = 0;
+
+ T const A = glm::nextPowerOfTwo(static_cast<T>(7));
+ Error += A == static_cast<T>(8) ? 0 : 1;
+
+ T const B = glm::nextPowerOfTwo(static_cast<T>(15));
+ Error += B == static_cast<T>(16) ? 0 : 1;
+
+ T const C = glm::nextPowerOfTwo(static_cast<T>(31));
+ Error += C == static_cast<T>(32) ? 0 : 1;
+
+ T const D = glm::nextPowerOfTwo(static_cast<T>(32));
+ Error += D == static_cast<T>(32) ? 0 : 1;
+
+ return Error;
+ }
+
+ int test()
+ {
+ int Error = 0;
+
+ Error += run<glm::int8>();
+ Error += run<glm::int16>();
+ Error += run<glm::int32>();
+ Error += run<glm::int64>();
+
+ Error += run<glm::uint8>();
+ Error += run<glm::uint16>();
+ Error += run<glm::uint32>();
+ Error += run<glm::uint64>();
+
+ return Error;
+ }
+}//namespace nextPowerOfTwo
+
+namespace prevMultiple
+{
+ template<typename genIUType>
+ struct type
+ {
+ genIUType Source;
+ genIUType Multiple;
+ genIUType Return;
+ };
+
+ template <typename T>
+ int run()
+ {
+ type<T> const Data[] =
+ {
+ {8, 3, 6},
+ {7, 7, 7}
+ };
+
+ int Error = 0;
+
+ for(std::size_t i = 0, n = sizeof(Data) / sizeof(type<T>); i < n; ++i)
+ {
+ T const Result = glm::prevMultiple(Data[i].Source, Data[i].Multiple);
+ Error += Data[i].Return == Result ? 0 : 1;
+ }
+
+ return Error;
+ }
+
+ int test()
+ {
+ int Error = 0;
+
+ Error += run<glm::int8>();
+ Error += run<glm::int16>();
+ Error += run<glm::int32>();
+ Error += run<glm::int64>();
+
+ Error += run<glm::uint8>();
+ Error += run<glm::uint16>();
+ Error += run<glm::uint32>();
+ Error += run<glm::uint64>();
+
+ return Error;
+ }
+}//namespace prevMultiple
+
+namespace nextMultiple
+{
+ static glm::uint const Multiples = 128;
+
+ int perf_nextMultiple(glm::uint Samples)
+ {
+ std::vector<glm::uint> Results(Samples * Multiples);
+
+ std::chrono::high_resolution_clock::time_point t0 = std::chrono::high_resolution_clock::now();
+
+ for(glm::uint Source = 0; Source < Samples; ++Source)
+ for(glm::uint Multiple = 0; Multiple < Multiples; ++Multiple)
+ {
+ Results[Source * Multiples + Multiple] = glm::nextMultiple(Source, Multiples);
+ }
+
+ std::chrono::high_resolution_clock::time_point t1 = std::chrono::high_resolution_clock::now();
+
+ std::printf("- glm::nextMultiple Time %d microseconds\n", static_cast<int>(std::chrono::duration_cast<std::chrono::microseconds>(t1 - t0).count()));
+
+ glm::uint Result = 0;
+ for(std::size_t i = 0, n = Results.size(); i < n; ++i)
+ Result += Results[i];
+
+ return Result > 0 ? 0 : 1;
+ }
+
+ template <typename T>
+ GLM_FUNC_QUALIFIER T nextMultipleMod(T Source, T Multiple)
+ {
+ T const Tmp = Source - static_cast<T>(1);
+ return Tmp + (Multiple - (Tmp % Multiple));
+ }
+
+ int perf_nextMultipleMod(glm::uint Samples)
+ {
+ std::vector<glm::uint> Results(Samples * Multiples);
+
+ std::chrono::high_resolution_clock::time_point t0 = std::chrono::high_resolution_clock::now();
+
+ for(glm::uint Multiple = 0; Multiple < Multiples; ++Multiple)
+ for (glm::uint Source = 0; Source < Samples; ++Source)
+ {
+ Results[Source * Multiples + Multiple] = nextMultipleMod(Source, Multiples);
+ }
+
+ std::chrono::high_resolution_clock::time_point t1 = std::chrono::high_resolution_clock::now();
+
+ std::printf("- nextMultipleMod Time %d microseconds\n", static_cast<int>(std::chrono::duration_cast<std::chrono::microseconds>(t1 - t0).count()));
+
+ glm::uint Result = 0;
+ for(std::size_t i = 0, n = Results.size(); i < n; ++i)
+ Result += Results[i];
+
+ return Result > 0 ? 0 : 1;
+ }
+
+ template <typename T>
+ GLM_FUNC_QUALIFIER T nextMultipleNeg(T Source, T Multiple)
+ {
+ if(Source > static_cast<T>(0))
+ {
+ T const Tmp = Source - static_cast<T>(1);
+ return Tmp + (Multiple - (Tmp % Multiple));
+ }
+ else
+ return Source + (-Source % Multiple);
+ }
+
+ int perf_nextMultipleNeg(glm::uint Samples)
+ {
+ std::vector<glm::uint> Results(Samples * Multiples);
+
+ std::chrono::high_resolution_clock::time_point t0 = std::chrono::high_resolution_clock::now();
+
+ for(glm::uint Source = 0; Source < Samples; ++Source)
+ for(glm::uint Multiple = 0; Multiple < Multiples; ++Multiple)
+ {
+ Results[Source * Multiples + Multiple] = nextMultipleNeg(Source, Multiples);
+ }
+
+ std::chrono::high_resolution_clock::time_point t1 = std::chrono::high_resolution_clock::now();
+
+ std::printf("- nextMultipleNeg Time %d microseconds\n", static_cast<int>(std::chrono::duration_cast<std::chrono::microseconds>(t1 - t0).count()));
+
+ glm::uint Result = 0;
+ for (std::size_t i = 0, n = Results.size(); i < n; ++i)
+ Result += Results[i];
+
+ return Result > 0 ? 0 : 1;
+ }
+
+ template <typename T>
+ GLM_FUNC_QUALIFIER T nextMultipleUFloat(T Source, T Multiple)
+ {
+ return Source + (Multiple - std::fmod(Source, Multiple));
+ }
+
+ int perf_nextMultipleUFloat(glm::uint Samples)
+ {
+ std::vector<float> Results(Samples * Multiples);
+
+ std::chrono::high_resolution_clock::time_point t0 = std::chrono::high_resolution_clock::now();
+
+ for(glm::uint Source = 0; Source < Samples; ++Source)
+ for(glm::uint Multiple = 0; Multiple < Multiples; ++Multiple)
+ {
+ Results[Source * Multiples + Multiple] = nextMultipleUFloat(static_cast<float>(Source), static_cast<float>(Multiples));
+ }
+
+ std::chrono::high_resolution_clock::time_point t1 = std::chrono::high_resolution_clock::now();
+
+ std::printf("- nextMultipleUFloat Time %d microseconds\n", static_cast<int>(std::chrono::duration_cast<std::chrono::microseconds>(t1 - t0).count()));
+
+ float Result = 0;
+ for (std::size_t i = 0, n = Results.size(); i < n; ++i)
+ Result += Results[i];
+
+ return Result > 0.0f ? 0 : 1;
+ }
+
+ template <typename T>
+ GLM_FUNC_QUALIFIER T nextMultipleFloat(T Source, T Multiple)
+ {
+ if(Source > static_cast<float>(0))
+ return Source + (Multiple - std::fmod(Source, Multiple));
+ else
+ return Source + std::fmod(-Source, Multiple);
+ }
+
+ int perf_nextMultipleFloat(glm::uint Samples)
+ {
+ std::vector<float> Results(Samples * Multiples);
+
+ std::chrono::high_resolution_clock::time_point t0 = std::chrono::high_resolution_clock::now();
+
+ for(glm::uint Source = 0; Source < Samples; ++Source)
+ for(glm::uint Multiple = 0; Multiple < Multiples; ++Multiple)
+ {
+ Results[Source * Multiples + Multiple] = nextMultipleFloat(static_cast<float>(Source), static_cast<float>(Multiples));
+ }
+
+ std::chrono::high_resolution_clock::time_point t1 = std::chrono::high_resolution_clock::now();
+
+ std::printf("- nextMultipleFloat Time %d microseconds\n", static_cast<int>(std::chrono::duration_cast<std::chrono::microseconds>(t1 - t0).count()));
+
+ float Result = 0;
+ for (std::size_t i = 0, n = Results.size(); i < n; ++i)
+ Result += Results[i];
+
+ return Result > 0.0f ? 0 : 1;
+ }
+
+ template<typename genIUType>
+ struct type
+ {
+ genIUType Source;
+ genIUType Multiple;
+ genIUType Return;
+ };
+
+ template <typename T>
+ int test_uint()
+ {
+ type<T> const Data[] =
+ {
+ { 3, 4, 4 },
+ { 6, 3, 6 },
+ { 5, 3, 6 },
+ { 7, 7, 7 },
+ { 0, 1, 0 },
+ { 8, 3, 9 }
+ };
+
+ int Error = 0;
+
+ for(std::size_t i = 0, n = sizeof(Data) / sizeof(type<T>); i < n; ++i)
+ {
+ T const Result0 = glm::nextMultiple(Data[i].Source, Data[i].Multiple);
+ Error += Data[i].Return == Result0 ? 0 : 1;
+ assert(!Error);
+
+ T const Result1 = nextMultipleMod(Data[i].Source, Data[i].Multiple);
+ Error += Data[i].Return == Result1 ? 0 : 1;
+ assert(!Error);
+ }
+
+ return Error;
+ }
+
+ int perf()
+ {
+ int Error = 0;
+
+ glm::uint const Samples = 10000;
+
+ for(int i = 0; i < 4; ++i)
+ {
+ std::printf("Run %d :\n", i);
+ Error += perf_nextMultiple(Samples);
+ Error += perf_nextMultipleMod(Samples);
+ Error += perf_nextMultipleNeg(Samples);
+ Error += perf_nextMultipleUFloat(Samples);
+ Error += perf_nextMultipleFloat(Samples);
+ std::printf("\n");
+ }
+
+ return Error;
+ }
+
+ int test()
+ {
+ int Error = 0;
+
+ Error += test_uint<glm::int8>();
+ Error += test_uint<glm::int16>();
+ Error += test_uint<glm::int32>();
+ Error += test_uint<glm::int64>();
+
+ Error += test_uint<glm::uint8>();
+ Error += test_uint<glm::uint16>();
+ Error += test_uint<glm::uint32>();
+ Error += test_uint<glm::uint64>();
+
+ return Error;
+ }
+}//namespace nextMultiple
+
+namespace findNSB
+{
+ template<typename T>
+ struct type
+ {
+ T Source;
+ int SignificantBitCount;
+ int Return;
+ };
+
+ template <typename T>
+ int run()
+ {
+ type<T> const Data[] =
+ {
+ { 0x00, 1,-1 },
+ { 0x01, 2,-1 },
+ { 0x02, 2,-1 },
+ { 0x06, 3,-1 },
+ { 0x01, 1, 0 },
+ { 0x03, 1, 0 },
+ { 0x03, 2, 1 },
+ { 0x07, 2, 1 },
+ { 0x05, 2, 2 },
+ { 0x0D, 2, 2 }
+ };
+
+ int Error = 0;
+
+ for (std::size_t i = 0, n = sizeof(Data) / sizeof(type<T>); i < n; ++i)
+ {
+ int const Result0 = glm::findNSB(Data[i].Source, Data[i].SignificantBitCount);
+ Error += Data[i].Return == Result0 ? 0 : 1;
+ assert(!Error);
+ }
+
+ return Error;
+ }
+
+ int test()
+ {
+ int Error = 0;
+
+ Error += run<glm::uint8>();
+ Error += run<glm::uint16>();
+ Error += run<glm::uint32>();
+ Error += run<glm::uint64>();
+
+ Error += run<glm::int8>();
+ Error += run<glm::int16>();
+ Error += run<glm::int32>();
+ Error += run<glm::int64>();
+
+ return Error;
+ }
+}//namespace findNSB
+
+int main()
+{
+ int Error = 0;
+
+ Error += findNSB::test();
+
+ Error += isPowerOfTwo::test();
+ Error += prevPowerOfTwo::test();
+ Error += nextPowerOfTwo::test();
+ Error += nextPowerOfTwo_advanced::test();
+ Error += prevMultiple::test();
+ Error += nextMultiple::test();
+
+# ifdef NDEBUG
+ Error += nextPowerOfTwo_advanced::perf();
+ Error += nextMultiple::perf();
+# endif//NDEBUG
+
+ return Error;
+}
+
+#else
+
+int main()
+{
+ return 0;
+}
+
+#endif