math: add sqrt_arm.s and sqrtGoC.go as fallback to soft fp emulation

5a: add SQRTF and SQRTD
5l: add ASQRTF and ASQRTD

Use ARMv7 VFP VSQRT instruction to speed up math.Sqrt

R=rsc, dave, m
CC=golang-dev
https://golang.org/cl/4551082

src/cmd/5a/lex.c

@@ -338,6 +338,8 @@ struct
 	"NRMD",		LTYPEI,	ANRMD,
 */
 
+	"SQRTF",	LTYPEI, ASQRTF,
+	"SQRTD",	LTYPEI, ASQRTD,
 	"CMPF",		LTYPEL, ACMPF,
 	"CMPD",		LTYPEL, ACMPD,
 	"ADDF",		LTYPEK,	AADDF,

src/cmd/5l/5.out.h

@@ -126,6 +126,8 @@ enum	as
 	AMULD,
 	ADIVF,
 	ADIVD,
+	ASQRTF,
+	ASQRTD,
 
 	ASRL,
 	ASRA,

src/cmd/5l/asm.c

@@ -1186,7 +1186,7 @@ if(debug['G']) print("%ux: %s: arm %d %d %d\n", (uint32)(p->pc), p->from.sym->na
 		r = p->reg;
 		if(r == NREG) {
 			r = rt;
-			if(p->as == AMOVF || p->as == AMOVD)
+			if(p->as == AMOVF || p->as == AMOVD || p->as == ASQRTF || p->as == ASQRTD)
 				r = 0;
 		}
 		o1 |= rf | (r<<16) | (rt<<12);
@@ -1620,6 +1620,8 @@ oprrr(int a, int sc)
 	case AMULF:	return o | (0xe<<24) | (0x2<<20) | (0xa<<8) | (0<<4);
 	case ADIVD:	return o | (0xe<<24) | (0x8<<20) | (0xb<<8) | (0<<4);
 	case ADIVF:	return o | (0xe<<24) | (0x8<<20) | (0xa<<8) | (0<<4);
+	case ASQRTD:	return o | (0xe<<24) | (0xb<<20) | (1<<16) | (0xb<<8) | (0xc<<4);
+	case ASQRTF:	return o | (0xe<<24) | (0xb<<20) | (1<<16) | (0xa<<8) | (0xc<<4);
 	case ACMPD:	return o | (0xe<<24) | (0xb<<20) | (4<<16) | (0xb<<8) | (0xc<<4);
 	case ACMPF:	return o | (0xe<<24) | (0xb<<20) | (4<<16) | (0xa<<8) | (0xc<<4);
 

src/cmd/5l/obj.c

@@ -67,6 +67,7 @@ static char*
 linkername[] =
 {
 	"runtime.softfloat",
+	"math.sqrtGoC",
 };
 
 void

src/cmd/5l/softfloat.c

@@ -54,6 +54,8 @@ softfloat(void)
 			case AMULD:
 			case ADIVF:
 			case ADIVD:
+			case ASQRTF:
+			case ASQRTD:
 				goto soft;
 
 			default:

src/cmd/5l/span.c

@@ -1023,6 +1023,8 @@ buildop(void)
 			oprange[AMULD] = oprange[r];
 			oprange[ADIVF] = oprange[r];
 			oprange[ADIVD] = oprange[r];
+			oprange[ASQRTF] = oprange[r];
+			oprange[ASQRTD] = oprange[r];
 			oprange[AMOVFD] = oprange[r];
 			oprange[AMOVDF] = oprange[r];
 			break;

src/pkg/math/Makefile

@@ -6,6 +6,9 @@ include ../../Make.inc
 
 TARG=math
 
+OFILES_arm=\
+	sqrt_arm.$O\
+
 OFILES_amd64=\
 	exp_amd64.$O\
 	fabs_amd64.$O\

src/pkg/math/sqrt_arm.s

@@ -0,0 +1,10 @@
+// Copyright 2011 The Go Authors.  All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+// func Sqrt(x float64) float64	
+TEXT ·Sqrt(SB),7,$0
+	MOVD   x+0(FP),F0
+	SQRTD  F0,F0
+	MOVD  F0,r+8(FP)
+	RET

src/pkg/math/sqrt_port.go

@@ -141,3 +141,7 @@ func sqrtGo(x float64) float64 {
 	ix = q>>1 + uint64(exp-1+bias)<<shift // significand + biased exponent
 	return Float64frombits(ix)
 }
+
+func sqrtGoC(f float64, r *float64) {
+	*r = sqrtGo(f)
+}

src/pkg/runtime/arm/softfloat.c

@@ -357,6 +357,15 @@ stage3:	// regd, regm are 4bit variables
 				regd, regm, m->freghi[regd], m->freglo[regd]);
 		break;
 
+	case 0xeeb10bc0:	// D[regd] = sqrt D[regm]
+		math·sqrtGoC(getd(regm), &uval);
+		putd(regd, uval);
+
+		if(trace)
+			runtime·printf("*** D[%d] = sqrt D[%d] %x-%x\n",
+				regd, regm, m->freghi[regd], m->freglo[regd]);
+		break;
+
 	case 0xeeb40bc0:	// D[regd] :: D[regm] (CMPD)
 		runtime·fcmp64c(getd(regd), getd(regm), &cmp, &nan);
 		m->fflag = fstatus(nan, cmp);