prevent statement reordering

nustxujun · Jun 5, 2024 · 90a8bbf · 90a8bbf
1 parent 9085daa
commit 90a8bbf
Show file tree

Hide file tree

Showing 3 changed files with 99 additions and 53 deletions.
diff --git a/README.md b/README.md
@@ -46,29 +46,8 @@ fixed64 is designed to find a balance between percision and performance. It is s
 - support speed up the multiplication and division with integer
 
 ### Performance
-- forceinline 
-- no overflow 
-- no hardware int128
 
-Intel Core i9-12900K 3.2GHz
-
-|Arithmetic|Fixed64|Hardware Float|
-|-|:-:|:-:|
-|Addition/Subtraction|0.027 ns|0.433 ns|
-|Multiplication|2.621 ns|0.837 ns|
-|Division|1.316 ns|2.784 ns|
-
-Apple M1 pro
-
-|Arithmetic|Fixed64|Hardware Float|
-|-|:-:|:-:|
-|Addition/Subtraction|0.000001 ns*|0.953 ns|
-|Multiplication|4.057 ns|1.246 ns|
-|Division|1.102 ns|3.144 ns|
-
-    * result is calculated with random operand, can not be calculated in compile time. 
-
-see more in ``benchmark.cpp``
+see more in [benchmark](https://github.com/nustxujun/FixedPoint64/blob/main/benchmark/benchmark.md)
 ### Supported Switcher
 ```c++
 #define FIXED_64_ENABLE_ROUNDING // apply rounding 
@@ -131,31 +110,9 @@ fixed64综合考虑了精度与性能的问题，使用了int64存储。个人
 - 支持与整型的乘除法加速
 
 ### Performance
-- 开启强制内敛 
-- 无溢出检测
-- 无硬件int128支持
-
-
-Intel Core i9-12900K 3.2GHz
-
-|算数操作|定点数|系统浮点数|
-|-|:-:|:-:|
-|加/减|0.027 ns|0.433 ns|
-|乘|2.621 ns|0.837 ns|
-|除|1.316 ns|2.784 ns|
-
-
-Apple M1 pro
-
-|算数操作|定点数|系统浮点数|
-|-|:-:|:-:|
-|加/减|0.000001 ns*|0.953 ns|
-|乘|4.057 ns|1.246 ns|
-|除|1.102 ns|3.144 ns|
 
-    * 计算数值是随机的，不可能是编译期计算出来的
+具体参考[benchmark](https://github.com/nustxujun/FixedPoint64/blob/main/benchmark/benchmark.md)
 
-具体请参考``benchmark.cpp``
 ### 开关
 ```c++
 #define FIXED_64_ENABLE_ROUNDING // 使用四舍五入

diff --git a/benchmark/benchmark.cpp b/benchmark/benchmark.cpp
@@ -40,9 +40,9 @@ struct Operand
 	fixed fa;
 	fixed fb;
 
-	inline Operand(float Min, float Max)
+	inline Operand(fp Min, fp Max)
 	{
-		std::uniform_real_distribution<fp> u(Min, Max);
+		std::uniform_real_distribution<double> u(Min, Max);
 		a = u(e) ;
 		b = u(e);
 		fa = a;
@@ -84,13 +84,27 @@ FIXED_64_FORCEINLINE void PreventOptimizedAway(fixed val)
 		EXPR1;\
 	}
 
-#pragma optimize("",off) // prevent statement reordering
+// prevent statment reordering
+#ifdef _MSC_VER
+#pragma optimize("",off) 
+#elif defined(__clang__)
+#pragma clang optimize off
+#else
+#pragma GCC push_options
+#pragma GCC optimize ("O0")
+#endif
 template<class T>
 void run_test(T& a, T& b, std::function<void(T&, T&)>&& f)
 {
 	f(a,b);
 }
-#pragma optimize("",on)
+#ifdef _MSC_VER
+#pragma optimize("",on) 
+#elif defined(__clang__)
+#pragma clang optimize on
+#else
+#pragma GCC pop_options
+#endif
 
 
 #define RUN_TEST(EXPR1, EXPR2, COUNT, Min, Max) \
@@ -114,21 +128,25 @@ struct TestGroup
 	std::string name;
 	uint64_t num_batch;
 	uint64_t count;
+	fp min;
+	fp max;
 	TestGroup(std::string n, uint64_t num, uint64_t c, fp min, fp max)
 	{
+		this->min = min;
+		this->max = max;
 		name = n;
 		num_batch = num;
 		count = c;
 		totals[0] = 0;
 		totals[1] = 0;
-		printf("%s [%f, %f]\n", name.c_str(), min, max);
 	}
 
 	~TestGroup()
 	{
-		printf("hard float: %lf ns, fixed point: %lf ns\n\n",
-			double(totals[0]) /count / num_batch
-			,double(totals[1]) / count / num_batch
+		printf("%16s[%6.1f, %6.1f]| %3.4lf ns | %3.4lf ns |\n",
+			name.c_str(),(float)min, (float)max,
+			double(totals[1]) /count / num_batch
+			,double(totals[0]) / count / num_batch
 			);
 	}
 };
@@ -167,6 +185,8 @@ auto benchmark = [](){
 	const uint64_t count1 = 0xffff'ff;
 	const uint64_t count2 = 0xffff'f;
 
+	printf("           arithmetic[ min, max]|fixed point| hard float|\n");
+
 	RUN_BASIC_TEST_GROUP("add/sub", +, -, 0xff, count1, -100, 100);
 
 #if FIXED_64_ENABLE_INT128_ACCELERATION

diff --git a/benchmark/benchmark.md b/benchmark/benchmark.md
@@ -0,0 +1,69 @@
+## Intel  Core i9-12900K 3.2GHz windows11 clang c++20
+
+|Arithmetic|Fixed64|Hardware Float|
+|:-|-:|-:|
+ add/sub[-100.0,  100.0]|0.0038 ns|0.4295 ns|
+mul[-100.0,  100.0]|2.6078 ns|0.9425 ns|
+mul[   0.0,    0.5]|2.9100 ns|0.9226 ns|
+mul[   0.5,    1.0]|2.6604 ns|27.1776 ns|
+mul[   1.0,    2.0]|2.6227 ns|0.8468 ns|
+mul[   2.0,  100.0]|2.6320 ns|0.8465 ns|
+div[-100.0,  100.0]|1.2967 ns|2.7318 ns|
+div[   0.0,    0.5]|12.0082 ns|2.3092 ns|
+div[   0.5,    1.0]|11.9341 ns|2.3041 ns|
+div[   1.0,    2.0]|5.8836 ns|29.7369 ns|
+div[   2.0,  100.0]|1.1417 ns|2.2860 ns|
+ceil[  -2.0,    2.0]|0.3916 ns|1.9973 ns|
+floor[  -2.0,    2.0]|0.3870 ns|1.7243 ns|
+round[  -2.0,    2.0]|0.6668 ns|6.5249 ns|
+abs[  -2.0,    2.0]|0.0538 ns|0.4180 ns|
+exp[   0.0,    1.0]|3.6085 ns|2.1249 ns|
+exp2[   0.0,    1.0]|2.7666 ns|30.3101 ns|
+sqrt[   0.0,  100.0]|13.7207 ns|1.6524 ns|
+sin[ -10.0,   10.0]|1.0053 ns|3.6804 ns|
+cos[ -10.0,   10.0]|1.3183 ns|3.4292 ns|
+tan[ -10.0,   10.0]|7.3048 ns|3.5210 ns|
+asin[  -1.0,    1.0]|23.3574 ns|2.7947 ns|
+acos[  -1.0,    1.0]|22.9430 ns|2.8789 ns|
+atan[   1.0,  100.0]|6.9723 ns|3.0697 ns|
+atan[   1.0,  100.0]|6.9567 ns|3.4708 ns|
+
+
+## Intel  Core i9-12900K 3.2GHz windows11 MSVC c++20
+
+|Arithmetic|Fixed64|Hardware Float|
+|:-|-:|-:|
+add/sub[-100.0,  100.0]| 0.2107 ns | 0.4187 ns |
+mul[-100.0,  100.0]| 2.7356 ns | 0.8556 ns |
+mul[   0.0,    0.5]| 2.6802 ns | 0.9478 ns |
+mul[   0.5,    1.0]| 2.6930 ns | 0.8574 ns |
+mul[   1.0,    2.0]| 2.6725 ns | 0.8455 ns |
+mul[   2.0,  100.0]| 2.6829 ns | 0.8512 ns |
+div[-100.0,  100.0]| 1.6081 ns | 3.1777 ns |
+div[   0.0,    0.5]| 13.7667 ns | 28.6761 ns |
+div[   0.5,    1.0]| 13.7861 ns | 29.7010 ns |
+div[   1.0,    2.0]| 6.2241 ns | 29.6876 ns |
+div[   2.0,  100.0]| 1.5079 ns | 2.3074 ns |
+ceil[  -2.0,    2.0]| 0.4330 ns | 2.0436 ns |
+floor[  -2.0,    2.0]| 0.4341 ns | 2.0402 ns |
+round[  -2.0,    2.0]| 2.0854 ns | 6.4961 ns |
+abs[  -2.0,    2.0]| 0.2149 ns | 0.4175 ns |
+exp[   0.0,    1.0]| 10.8654 ns | 2.0519 ns |
+exp2[   0.0,    1.0]| 9.4355 ns | 30.2611 ns |
+sqrt[   0.0,  100.0]| 10.9269 ns | 0.6271 ns |
+sin[ -10.0,   10.0]| 2.3128 ns | 3.9287 ns |
+cos[ -10.0,   10.0]| 2.1791 ns | 3.6846 ns |
+tan[ -10.0,   10.0]| 12.0050 ns | 3.6412 ns |
+asin[  -1.0,    1.0]| 19.5066 ns | 2.7914 ns |
+acos[  -1.0,    1.0]| 20.2079 ns | 2.9695 ns |
+atan[   1.0,  100.0]| 17.5488 ns | 3.1872 ns |
+atan[   1.0,  100.0]| 17.1658 ns | 3.0822 ns |
+
+
+## Apple M1 pro
+
+|Arithmetic|Fixed64|Hardware Float|
+|-|-:|-:|
+|Addition/Subtraction|0.215 ns|0.953 ns|
+|Multiplication|4.057 ns|1.246 ns|
+|Division|1.102 ns|3.144 ns|