cmd/asm: add s390x branch-on-count instructions

The branch-on-count instructions on s390x decrement the input
register and then compare its value to 0. If not equal the branch
is taken.

These instructions are useful for implementing loops with a set
number of iterations (which might be in a register).

For example, this for loop:

	for i := 0; i < n; i++ {
		... // i is not used or modified in the loop
	}

Could be implemented using this assembly:

	MOVD  Rn, Ri
loop:
	...
	BRCTG Ri, loop

Note that i will count down from n in the assembly whereas in the
original for loop it counted up to n which is why we can't use i
in the loop.

These instructions will only be used in hand-written codegen and
assembly for now since SSA blocks cannot currently modify values.
We could look into this in the future though.

Change-Id: Iaab93b8aa2699513b825439b8ea20d8fe2ea1ee6
Reviewed-on: https://go-review.googlesource.com/c/go/+/199977
Run-TryBot: Michael Munday <[email protected]>
TryBot-Result: Gobot Gobot <[email protected]>
Reviewed-by: Cherry Zhang <[email protected]>

src/cmd/asm/internal/arch/s390x.go

@@ -30,6 +30,8 @@ func jumpS390x(word string) bool {
 		"BR",
 		"BVC",
 		"BVS",
+		"BRCT",
+		"BRCTG",
 		"CMPBEQ",
 		"CMPBGE",
 		"CMPBGT",

src/cmd/asm/internal/asm/testdata/s390x.s

@@ -266,6 +266,9 @@ TEXT main·foo(SB),DUPOK|NOSPLIT,$16-0 // TEXT main.foo(SB), DUPOK|NOSPLIT, $16-
 	BLTU	0(PC)                  // a7540000
 	BLEU	0(PC)                  // a7d40000
 
+	BRCT	R1, 0(PC)              // a7160000
+	BRCTG	R2, 0(PC)              // a7270000
+
 	CMPBNE	R1, R2, 0(PC)          // ec1200007064
 	CMPBEQ	R3, R4, 0(PC)          // ec3400008064
 	CMPBLT	R5, R6, 0(PC)          // ec5600004064

src/cmd/compile/internal/s390x/ggen.go

@@ -38,18 +38,14 @@ func zerorange(pp *gc.Progs, p *obj.Prog, off, cnt int64, _ *uint32) *obj.Prog {
 
 	// Generate a loop of large clears.
 	if cnt > clearLoopCutoff {
-		n := cnt - (cnt % 256)
-		end := int16(s390x.REGRT2)
-		p = pp.Appendpp(p, s390x.AADD, obj.TYPE_CONST, 0, off+n, obj.TYPE_REG, end, 0)
-		p.Reg = reg
+		ireg := int16(s390x.REGRT2) // register holds number of remaining loop iterations
+		p = pp.Appendpp(p, s390x.AMOVD, obj.TYPE_CONST, 0, cnt/256, obj.TYPE_REG, ireg, 0)
 		p = pp.Appendpp(p, s390x.ACLEAR, obj.TYPE_CONST, 0, 256, obj.TYPE_MEM, reg, off)
 		pl := p
 		p = pp.Appendpp(p, s390x.AADD, obj.TYPE_CONST, 0, 256, obj.TYPE_REG, reg, 0)
-		p = pp.Appendpp(p, s390x.ACMP, obj.TYPE_REG, reg, 0, obj.TYPE_REG, end, 0)
-		p = pp.Appendpp(p, s390x.ABNE, obj.TYPE_NONE, 0, 0, obj.TYPE_BRANCH, 0, 0)
+		p = pp.Appendpp(p, s390x.ABRCTG, obj.TYPE_REG, ireg, 0, obj.TYPE_BRANCH, 0, 0)
 		gc.Patch(p, pl)
-
-		cnt -= n
+		cnt = cnt % 256
 	}
 
 	// Generate remaining clear instructions without a loop.

src/cmd/internal/obj/s390x/a.out.go

@@ -409,6 +409,10 @@ const (
 	ABVS
 	ASYSCALL
 
+	// branch on count
+	ABRCT
+	ABRCTG
+
 	// compare and branch
 	ACRJ
 	ACGRJ

src/cmd/internal/obj/s390x/asmz.go

@@ -256,6 +256,10 @@ var optab = []Optab{
 	{i: 90, as: ACLGIJ, a1: C_SCON, a2: C_REG, a3: C_ADDCON, a6: C_SBRA},
 	{i: 90, as: ACMPUBEQ, a1: C_REG, a3: C_ANDCON, a6: C_SBRA},
 
+	// branch on count
+	{i: 41, as: ABRCT, a1: C_REG, a6: C_SBRA},
+	{i: 41, as: ABRCTG, a1: C_REG, a6: C_SBRA},
+
 	// move on condition
 	{i: 17, as: AMOVDEQ, a1: C_REG, a6: C_REG},
 
@@ -3394,6 +3398,21 @@ func (c *ctxtz) asmout(p *obj.Prog, asm *[]byte) {
 			*asm = append(*asm, uint8(wd))
 		}
 
+	case 41: // branch on count
+		r1 := p.From.Reg
+		ri2 := (p.Pcond.Pc - p.Pc) >> 1
+		if int64(int16(ri2)) != ri2 {
+			c.ctxt.Diag("branch target too far away")
+		}
+		var opcode uint32
+		switch p.As {
+		case ABRCT:
+			opcode = op_BRCT
+		case ABRCTG:
+			opcode = op_BRCTG
+		}
+		zRI(opcode, uint32(r1), uint32(ri2), asm)
+
 	case 47: // negate [reg] reg
 		r := p.From.Reg
 		if r == 0 {