New scalar benchmark suite, comparing SkScalarAs2sCompliment vs native float operations

http://http://codereview.appspot.com/4532116/

git-svn-id: http://skia.googlecode.com/svn/trunk@1511 2bbb7eff-a529-9590-31e7-b0007b416f81

diff --git a/Makefile b/Makefile
index 31716746537c81559c05c9702a4093b4522f8b90..fcb5248ac88f71deda29520ddf959650fefcccdb 100644 (file)
--- a/Makefile
+++ b/Makefile
@@ -201,7 +201,7 @@ out/libskia.so: Makefile $(OBJ_LIST) $(JUST_COMPILE_OBJS)
 
 BENCH_SRCS := RectBench.cpp SkBenchmark.cpp benchmain.cpp BitmapBench.cpp \
                          RepeatTileBench.cpp DecodeBench.cpp FPSBench.cpp PathBench.cpp \
-                         GradientBench.cpp MatrixBench.cpp
+                         GradientBench.cpp MatrixBench.cpp ScalarBench.cpp
 
 BENCH_SRCS := $(addprefix bench/, $(BENCH_SRCS))
 

diff --git a/bench/ScalarBench.cpp b/bench/ScalarBench.cpp
new file mode 100644 (file)
index 0000000..639d10d
--- /dev/null
+++ b/bench/ScalarBench.cpp
@@ -0,0 +1,96 @@
+#include "SkBenchmark.h"
+#include "SkRandom.h"
+#include "SkString.h"
+
+class ScalarBench : public SkBenchmark {
+    SkString    fName;
+    enum { N = 100000 };
+public:
+    ScalarBench(void* param, const char name[]) : INHERITED(param) {
+        fName.printf("scalar_%s", name);
+    }
+
+    virtual void performTest() = 0;
+
+protected:
+    virtual int mulLoopCount() const { return 1; }
+
+    virtual const char* onGetName() {
+        return fName.c_str();
+    }
+
+    virtual void onDraw(SkCanvas* canvas) {
+        int n = N * this->mulLoopCount();
+        for (int i = 0; i < n; i++) {
+            this->performTest();
+        }
+    }
+
+private:
+    typedef SkBenchmark INHERITED;
+};
+
+// we want to stop the compiler from eliminating code that it thinks is a no-op
+// so we have a non-static global we increment, hoping that will convince the
+// compiler to execute everything
+int gScalarBench_NonStaticGlobal;
+
+#define always_do(pred)                     \
+    do {                                    \
+        if (pred) {                         \
+            ++gScalarBench_NonStaticGlobal; \
+        }                                   \
+    } while (0)
+
+// having unknown values in our arrays can throw off the timing a lot, perhaps
+// handling NaN values is a lot slower. Anyway, this guy is just meant to put
+// reasonable values in our arrays.
+template <typename T> void init9(T array[9]) {
+    SkRandom rand;
+    for (int i = 0; i < 9; i++) {
+        array[i] = rand.nextSScalar1();
+    }
+}
+
+class FloatComparisonBench : public ScalarBench {
+public:
+    FloatComparisonBench(void* param) : INHERITED(param, "compare_float") {
+        init9(fArray);
+    }
+protected:
+    virtual int mulLoopCount() const { return 4; }
+    virtual void performTest() {
+        always_do(fArray[6] != 0.0f || fArray[7] != 0.0f || fArray[8] != 1.0f);
+        always_do(fArray[2] != 0.0f || fArray[5] != 0.0f);
+    }
+private:
+    float fArray[9];
+    typedef ScalarBench INHERITED;
+};
+
+class ForcedIntComparisonBench : public ScalarBench {
+public:
+    ForcedIntComparisonBench(void* param)
+        : INHERITED(param, "compare_forced_int") {
+        init9(fArray);
+    }
+protected:
+    virtual int mulLoopCount() const { return 4; }
+    virtual void performTest() {
+        always_do(SkScalarAs2sCompliment(fArray[6]) |
+                  SkScalarAs2sCompliment(fArray[7]) |
+                  (SkScalarAs2sCompliment(fArray[8]) - kPersp1Int));
+        always_do(SkScalarAs2sCompliment(fArray[2]) |
+                  SkScalarAs2sCompliment(fArray[5]));
+    }
+private:
+    static const int32_t kPersp1Int = 0x3f800000;
+    float fArray[9];
+    typedef ScalarBench INHERITED;
+};
+
+static SkBenchmark* S0(void* p) { return new FloatComparisonBench(p); }
+static SkBenchmark* S1(void* p) { return new ForcedIntComparisonBench(p); }
+
+static BenchRegistry gReg0(S0);
+static BenchRegistry gReg1(S1);