fix the 9 asm notes. add VFMA231[SP]D instructions.

3 files changed, 47 insertions, 26 deletions

diff --git a/dppd.s b/dppd.s
index 2b1833c..907a437 100644
--- a/dppd.s
+++ b/dppd.s
@@ -7,8 +7,8 @@ TEXT dppd(SB), 1, $0
 	MOVQ SP, AX
 	MOVLPD(8, rAX, rX0)	/* MOVLPD a+0(FP), X0 */
 	MOVHPD(16, rAX, rX0)	/* MOVHPD a+8(FP), X0 */
-	MOVLPD(32, rAX, rX1)	/* MOVLPD b+0(FP), X1 */
-	MOVHPD(40, rAX, rX1)	/* MOVHPD b+8(FP), X1*/
+	MOVLPD(32, rAX, rX1)	/* MOVLPD b+24(FP), X1 */
+	MOVHPD(40, rAX, rX1)	/* MOVHPD b+32(FP), X1*/
 	DPPD(rX1, rX0)		/* DPPD $0x31, X1, X0 */
 	RET
 
@@ -16,12 +16,12 @@ TEXT dppd3(SB), 1, $0
 	MOVQ SP, AX
 	MOVLPD(8, rAX, rX0)	/* MOVLPD a+0(FP), X0 */
 	MOVHPD(16, rAX, rX0)	/* MOVHPD a+8(FP), X0 */
-	MOVLPD(40, rAX, rX1)	/* MOVLPD b+0(FP), X1 */
-	MOVHPD(48, rAX, rX1)	/* MOVHPD b+8(FP), X1 */
+	MOVLPD(40, rAX, rX1)	/* MOVLPD b+32(FP), X1 */
+	MOVHPD(48, rAX, rX1)	/* MOVHPD b+40(FP), X1 */
 	DPPD(rX1, rX0)		/* DPPD $0x31, X1, X0 */
 	MOVSD one(SB), X1
 	MOVHPD(24, rAX, rX0)	/* MOVHPD a+16(FP), X0 */
-	MOVHPD(56, rAX, rX1)	/* MOVHPD b+16(FP), X1 */
+	MOVHPD(56, rAX, rX1)	/* MOVHPD b+48(FP), X1 */
 	DPPD(rX1, rX0)		/* DPPD $0x31, X1, X0 */
 	RET
 
@@ -38,18 +38,18 @@ TEXT Pt2b(SB), 1, $0
 TEXT hsubpd(SB), 1, $0
 	MOVQ SP, AX
 	MOVLPD(8, rAX, rX0)	/* MOVLPD a+0(FP), X0 */
-	MOVHPD(16, rAX, rX0)	/* MOVHPD b+0(FP), X0 */
+	MOVHPD(16, rAX, rX0)	/* MOVHPD b+8(FP), X0 */
 	HSUBPD(rX0, rX0)	/* HSUBPD X0, X0 */
 	RET
 
 TEXT xvec3(SB), 1, $0
 	MOVQ SP, AX
 	ADDQ $8, AX
-	MOVLPD(40, rAX, rX0)	/* MOVLPD b+0(FP), X0 */
+	MOVLPD(40, rAX, rX0)	/* MOVLPD b+32(FP), X0 */
 	MOVHPD(8, rAX, rX0)	/* MOVHPD a+0(FP), X0 */
 	MOVLPD(16, rAX, rX1)	/* MOVLPD a+8(FP), X1 */
-	MOVHPD(48, rAX, rX1)	/* MOVHPD b+8(FP), X1 */
-	MOVLPD(56, rAX, rX2)	/* MOVLPD b+16(FP), X2 */
+	MOVHPD(48, rAX, rX1)	/* MOVHPD b+40(FP), X1 */
+	MOVLPD(56, rAX, rX2)	/* MOVLPD b+48(FP), X2 */
 	MOVHPD(24, rAX, rX2)	/* MOVHPD a+16(FP), X2 */
 	MOVAPD X1, X3
 	MULPD X2, X3
@@ -69,3 +69,10 @@ TEXT xvec3(SB), 1, $0
 	XORPD X0, X0
 	MOVSD X0, 24(DI)
 	RET
+
+TEXT fma(SB), 1, $0
+	MOVSD a+0(FP), X0
+	MOVSD b+8(FP), X1
+	MOVSD c+16(FP), X2
+	VFMADD231SD(rX1, rX2, rX0)
+	RET
diff --git a/main.c b/main.c
index f30da53..fe51889 100644
--- a/main.c
+++ b/main.c
@@ -9,6 +9,7 @@ double dppd3(Point3, Point3);
 Point2 Pt2b(double, double, double);
 Point3 xvec3(Point3, Point3);
 double hsubpd(double, double);
+double fma(double, double, double);
 
 double
 fmin(double a, double b)
@@ -78,5 +79,10 @@ main(int argc, char *argv[])
 	t1 = nanosec();
 	print("crossvec3(%V, %V) = %V\ttook %lludns\n", p0t, p1t, pr, t1-t0);
 
+	t0 = nanosec();
+	r = fma(a, b, 21);
+	t1 = nanosec();
+	print("fma(%g, %g, 21) = %g\ttook %lludns\n", a, b, r, t1-t0);
+
 	exits(nil);
 }
diff --git a/sse.h b/sse.h
index b2ba3f4..ffad543 100644
--- a/sse.h
+++ b/sse.h
@@ -15,15 +15,33 @@
 #define rX5	5
 #define rX6	6
 
+#define VEX_m_0F	(1)
+#define VEX_m_0F38	(2)
+#define VEX_m_0F3A	(3)
+#define VEX_L_128	(0)
+#define VEX_L_256	(1)
+#define VEX_p_NO	(0)
+#define VEX_p_66	(1)
+#define VEX_p_F3	(2)
+#define VEX_p_F2	(3)
+
 #define OP(o, m, ro, rm)	WORD $0x0F66; BYTE $(o);	\
 			BYTE $(((m)<<6)|((ro)<<3)|(rm))
 #define OPi(o, m, ro, rm, i)	OP((o), (m), (ro), (rm));	\
 			BYTE $(i)
-#define OP4(o, m, ro, rm)	LONG $0x(o)0F66;		\
+#define OP4(o, m, ro, rm)	WORD $0x0F66; WORD $(o);	\
 			BYTE $(((m)<<6)|((ro)<<3)|(rm))
 #define OP4i(o, m, ro, rm, i)	OP4((o), (m), (ro), (rm));	\
 			BYTE $(i)
 
+#define VEX3(r, x, b, m, w, v, l, p)	BYTE $0xC4;				\
+				BYTE $(((~r)<<7)|((~x)<<6)|((~b)<<5)|(m));	\
+				BYTE $(((w)<<7)|((~v)<<3)|((l)<<2)|(p))
+#define VEX2(r, b, l, p)	BYTE $0xC5;					\
+			BYTE $(((~r)<<7)|((~v)<<3)|((l)<<2)|(p))
+#define VOP(o, m, ro, rm)	BYTE $(o);	\
+			BYTE $(((m)<<6)|((ro)<<3)|(rm))
+
 /* MOVLPD */
 //opcode = 660F12
 //modrm  = 01 000 000 [AX → X0] / 01 001 000 [AX → X1]
@@ -47,20 +65,10 @@
 //imm8   = 0011 0001
 #define DPPD(s, d) OP4i(0x413A, 0x3, (d), (s), 0x31)
 
+/* VFMADD231SD (128 bit) */
+#define VFMADD231SD(s0, s1, d) VEX3(0,0,0,VEX_m_0F38,1,(s0),VEX_L_128,VEX_p_66);	\
+			VOP(0xB9, 0x3, (d), (s1))
 
-#define VEX_m_0F	(1)
-#define VEX_m_0F38	(2)
-#define VEX_m_0F3A	(3)
-#define VEX_L_128	(0)
-#define VEX_L_256	(1)
-#define VEX_p_NO	(0)
-#define VEX_p_66	(1)
-#define VEX_p_F3	(2)
-#define VEX_p_F2	(3)
-
-#define VEX2(r, x, b, m, w, v, l, p)	BYTE $0xC5;			\
-				BYTE $(((~r)<<7)|((~x)<<6)|((~b)<<5)|(m));	\
-				BYTE $(((w)<<7)|((~v)<<3)|((l)<<2)|(p));
-
-#define VEX3(r, b, l, p)	BYTE $0xC4;	\
-			BYTE $(((~r)<<7)|((~v)<<3)|((l)<<2)|(p));
+/* VFMADD231PD (128 bit) */
+#define VFMADD231PD(s0, s1, d) VEX3(0,0,0,VEX_m_0F38,1,(s0),VEX_L_128,VEX_p_66);	\
+			VOP(0xB8, 0x3, (d), (s1))