cleaned things up and improved the organization a bit.

2 files changed, 188 insertions, 14 deletions

diff --git a/bench/main.c b/bench/main.c
index 9d00719..811c471 100644
--- a/bench/main.c
+++ b/bench/main.c
@@ -2,13 +2,61 @@
 #include <libc.h>
 #include <thread.h>
 #include <geometry.h>
-#include "b.h"
+#include "../bench9/b.h"
 
-double dppd(Point2, Point2);
-double dppda(Point2, Point2);
-double dppd3(Point3, Point3);
-double dppd3a(Point3, Point3);
-Point3 xvec3(Point3, Point3);
+double min(double, double);
+double dotvec2_sse4(Point2, Point2);
+double dotvec2_avx(Point2, Point2);
+double dotvec3_sse4(Point3, Point3);
+double dotvec3_avx(Point3, Point3);
+Point2 Pt2b(double, double, double);
+Point3 crossvec3_sse(Point3, Point3);
+double hsubpd(double, double);
+double fma(double, double, double);
+Point2 addpt2_avx(Point2, Point2);
+
+double
+fmin(double a, double b)
+{
+	return a<b? a: b;
+}
+
+double
+madd(double a, double b, double c)
+{
+	return a + b*c;
+}
+
+static void
+bmin(int fd)
+{
+	Bgr g;
+	B *b0, *b1;
+	double a, b;
+	int i;
+
+	benchinitgr(&g, "min");
+	b0 = benchadd(&g, "fmin");
+	b1 = benchadd(&g, "fmin_sse");
+
+	while(b0->n > 0 || b1->n > 0){
+		a = truerand()*frand();
+		b = truerand()*frand();
+
+		benchin(b0);
+		for(i = 0; i < 1e6; i++)
+			fmin(a, b);
+		benchout(b0);
+
+		benchin(b1);
+		for(i = 0; i < 1e6; i++)
+			min(a, b);
+		benchout(b1);
+	}
+
+	benchprintgr(&g, fd);
+	benchfreegr(&g);
+}
 
 static void
 bdotvec2(int fd)
@@ -20,7 +68,7 @@ bdotvec2(int fd)
 
 	benchinitgr(&g, "2d dot product");
 	b0 = benchadd(&g, "dotvec2");
-	b1 = benchadd(&g, "dotvec2_simd");
+	b1 = benchadd(&g, "dotvec2_sse4");
 	b2 = benchadd(&g, "dotvec2_avx");
 
 	while(b0->n > 0 || b1->n > 0){
@@ -34,12 +82,12 @@ bdotvec2(int fd)
 
 		benchin(b1);
 		for(i = 0; i < 1e6; i++)
-			dppd(a, b);
+			dotvec2_sse4(a, b);
 		benchout(b1);
 
 		benchin(b2);
 		for(i = 0; i < 1e6; i++)
-			dppda(a, b);
+			dotvec2_avx(a, b);
 		benchout(b2);
 	}
 
@@ -57,7 +105,7 @@ bdotvec3(int fd)
 
 	benchinitgr(&g, "3d dot product");
 	b0 = benchadd(&g, "dotvec3");
-	b1 = benchadd(&g, "dotvec3_simd");
+	b1 = benchadd(&g, "dotvec3_sse4");
 	b2 = benchadd(&g, "dotvec3_avx");
 
 	while(b0->n > 0 || b1->n > 0){
@@ -71,12 +119,12 @@ bdotvec3(int fd)
 
 		benchin(b1);
 		for(i = 0; i < 1e6; i++)
-			dppd3(a, b);
+			dotvec3_sse4(a, b);
 		benchout(b1);
 
 		benchin(b2);
 		for(i = 0; i < 1e6; i++)
-			dppd3a(a, b);
+			dotvec3_avx(a, b);
 		benchout(b2);
 	}
 
@@ -94,7 +142,7 @@ bcrossvec3(int fd)
 
 	benchinitgr(&g, "3d cross product");
 	b0 = benchadd(&g, "crossvec3");
-	b1 = benchadd(&g, "crossvec3_simd");
+	b1 = benchadd(&g, "crossvec3_sse");
 
 	while(b0->n > 0 || b1->n > 0){
 		a = Vec3(truerand()*frand(), truerand()*frand(), truerand()*frand());
@@ -107,7 +155,102 @@ bcrossvec3(int fd)
 
 		benchin(b1);
 		for(i = 0; i < 1e6; i++)
-			xvec3(a, b);
+			crossvec3_sse(a, b);
+		benchout(b1);
+	}
+
+	benchprintgr(&g, fd);
+	benchfreegr(&g);
+}
+
+static void
+bPt2(int fd)
+{
+	Bgr g;
+	B *b0, *b1;
+	double x, y, w;
+	int i;
+
+	benchinitgr(&g, "Pt2");
+	b0 = benchadd(&g, "Pt2");
+	b1 = benchadd(&g, "Pt2b");
+
+	while(b0->n > 0 || b1->n > 0){
+		x = truerand()*frand();
+		y = truerand()*frand();
+		w = truerand()*frand();
+
+		benchin(b0);
+		for(i = 0; i < 1e6; i++)
+			Pt2(x, y, w);
+		benchout(b0);
+
+		benchin(b1);
+		for(i = 0; i < 1e6; i++)
+			Pt2b(x, y, w);
+		benchout(b1);
+	}
+
+	benchprintgr(&g, fd);
+	benchfreegr(&g);
+}
+
+static void
+bfma(int fd)
+{
+	Bgr g;
+	B *b0, *b1;
+	double a, b, c;
+	int i;
+
+	benchinitgr(&g, "multiply + add");
+	b0 = benchadd(&g, "madd");
+	b1 = benchadd(&g, "fma_avx");
+
+	while(b0->n > 0 || b1->n > 0){
+		a = truerand()*frand();
+		b = truerand()*frand();
+		c = truerand()*frand();
+
+		benchin(b0);
+		for(i = 0; i < 1e6; i++)
+			madd(a, b, c);
+		benchout(b0);
+
+		benchin(b1);
+		for(i = 0; i < 1e6; i++)
+			fma(a, b, c);
+		benchout(b1);
+	}
+
+	benchprintgr(&g, fd);
+	benchfreegr(&g);
+}
+
+static void
+baddpt2(int fd)
+{
+	Bgr g;
+	B *b0, *b1;
+	Point2 a, b;
+	int i;
+
+	benchinitgr(&g, "2d point sum");
+	b0 = benchadd(&g, "addpt2");
+	b1 = benchadd(&g, "addpt2_avx");
+
+	while(b0->n > 0 || b1->n > 0){
+		a = Pt2(truerand()*frand(), truerand()*frand(), truerand()*frand());
+		b = Pt2(truerand()*frand(), truerand()*frand(), truerand()*frand());
+
+		benchin(b0);
+		for(i = 0; i < 1e6; i++)
+			addpt2(a, b);
+		benchout(b0);
+
+		benchin(b1);
+		for(i = 0; i < 1e6; i++)
+			addpt2_avx(a, b);
 		benchout(b1);
 	}
 
@@ -124,11 +267,19 @@ threadmain(int argc, char **argv)
 	if(benchwire(0) != 0)
 		fprint(2, "failed to wire: %r\n");
 
+	bmin(1);
+	bseparator(1);
 	bdotvec2(1);
 	bseparator(1);
 	bdotvec3(1);
 	bseparator(1);
 	bcrossvec3(1);
+	bseparator(1);
+	bPt2(1);
+	bseparator(1);
+	bfma(1);
+	bseparator(1);
+	baddpt2(1);
 
 	threadexitsall(nil);
 }
diff --git a/bench/mkfile b/bench/mkfile
new file mode 100644
index 0000000..e649008
--- /dev/null
+++ b/bench/mkfile
@@ -0,0 +1,23 @@
+</$objtype/mkfile
+
+TARG=bench9
+BIN=/$objtype/bin
+arch=`{echo __^$objtype^__}
+CFLAGS=$CFLAGS -D$arch -p
+
+HFILES=\
+	../bench9/b.h\
+	../regs.h\
+	../sse.h\
+	../avx.h\
+
+OFILES=\
+	../bench9/b.$O\
+	../bench9/b_$objtype.$O\
+	../min.$O\
+	../dppd.$O\
+	main.$O\
+
+default:V: all
+
+</sys/src/cmd/mkone