cmd/compile: add floating point min/max intrinsics on s390x

Add the VECTOR FP (MINIMUM|MAXIMUM) instructions to the assembler and
use them in the compiler to implement min and max.

Note: I've allowed floating point registers to be used with the single
element instructions (those with the W instead of V prefix) to allow
easier integration into the compiler.

Change-Id: I5f80a510bd248cf483cce95f1979bf63fbae7de6
Reviewed-on: https://go-review.googlesource.com/c/go/+/684715
Reviewed-by: Keith Randall <khr@golang.org>
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com>
Reviewed-by: Mark Freeman <mark@golang.org>
Reviewed-by: Keith Randall <khr@google.com>

4 files changed, 58 insertions, 1 deletions

diff --git a/src/cmd/internal/obj/s390x/a.out.go b/src/cmd/internal/obj/s390x/a.out.go
index 1a64370efa..dc715182f5 100644
--- a/src/cmd/internal/obj/s390x/a.out.go
+++ b/src/cmd/internal/obj/s390x/a.out.go
@@ -715,6 +715,14 @@ const (
 	AWFLNDB
 	AVFLPDB
 	AWFLPDB
+	AVFMAXDB
+	AWFMAXDB
+	AVFMAXSB
+	AWFMAXSB
+	AVFMINDB
+	AWFMINDB
+	AVFMINSB
+	AWFMINSB
 	AVFSQ
 	AVFSQDB
 	AWFSQDB
diff --git a/src/cmd/internal/obj/s390x/anames.go b/src/cmd/internal/obj/s390x/anames.go
index c0a0c401fa..a6f2820f85 100644
--- a/src/cmd/internal/obj/s390x/anames.go
+++ b/src/cmd/internal/obj/s390x/anames.go
@@ -438,6 +438,14 @@ var Anames = []string{
 	"WFLNDB",
 	"VFLPDB",
 	"WFLPDB",
+	"VFMAXDB",
+	"WFMAXDB",
+	"VFMAXSB",
+	"WFMAXSB",
+	"VFMINDB",
+	"WFMINDB",
+	"VFMINSB",
+	"WFMINSB",
 	"VFSQ",
 	"VFSQDB",
 	"WFSQDB",
diff --git a/src/cmd/internal/obj/s390x/asmz.go b/src/cmd/internal/obj/s390x/asmz.go
index 72d92abbaf..957222a155 100644
--- a/src/cmd/internal/obj/s390x/asmz.go
+++ b/src/cmd/internal/obj/s390x/asmz.go
@@ -441,6 +441,11 @@ var optab = []Optab{
 	{i: 119, as: AVERLLVG, a1: C_VREG, a2: C_VREG, a6: C_VREG},
 	{i: 119, as: AVERLLVG, a1: C_VREG, a6: C_VREG},
 
+	// VRR-c floating point min/max
+	{i: 128, as: AVFMAXDB, a1: C_SCON, a2: C_VREG, a3: C_VREG, a6: C_VREG},
+	{i: 128, as: AWFMAXDB, a1: C_SCON, a2: C_VREG, a3: C_VREG, a6: C_VREG},
+	{i: 128, as: AWFMAXDB, a1: C_SCON, a2: C_FREG, a3: C_FREG, a6: C_FREG},
+
 	// VRR-d
 	{i: 120, as: AVACQ, a1: C_VREG, a2: C_VREG, a3: C_VREG, a6: C_VREG},
 
@@ -1480,6 +1485,14 @@ func buildop(ctxt *obj.Link) {
 			opset(AVFMSDB, r)
 			opset(AWFMSDB, r)
 			opset(AVPERM, r)
+		case AVFMAXDB:
+			opset(AVFMAXSB, r)
+			opset(AVFMINDB, r)
+			opset(AVFMINSB, r)
+		case AWFMAXDB:
+			opset(AWFMAXSB, r)
+			opset(AWFMINDB, r)
+			opset(AWFMINSB, r)
 		case AKM:
 			opset(AKMC, r)
 			opset(AKLMD, r)
@@ -2636,6 +2649,8 @@ const (
 	op_VUPLL  uint32 = 0xE7D4 // 	VRR-a	VECTOR UNPACK LOGICAL LOW
 	op_VUPL   uint32 = 0xE7D6 // 	VRR-a	VECTOR UNPACK LOW
 	op_VMSL   uint32 = 0xE7B8 // 	VRR-d	VECTOR MULTIPLY SUM LOGICAL
+	op_VFMAX  uint32 = 0xE7EF // 	VRR-c	VECTOR FP MAXIMUM
+	op_VFMIN  uint32 = 0xE7EE // 	VRR-c	VECTOR FP MINIMUM
 
 	// added in z15
 	op_KDSA uint32 = 0xB93A // FORMAT_RRE        COMPUTE DIGITAL SIGNATURE AUTHENTICATION (KDSA)
@@ -4475,6 +4490,12 @@ func (c *ctxtz) asmout(p *obj.Prog, asm *[]byte) {
 			c.ctxt.Diag("padding byte register cannot be same as input or output register %v", p)
 		}
 		zRS(op_MVCLE, uint32(p.To.Reg), uint32(p.Reg), uint32(p.From.Reg), uint32(d2), asm)
+
+	case 128: // VRR-c floating point max/min
+		op, m4, _ := vop(p.As)
+		m5 := singleElementMask(p.As)
+		m6 := uint32(c.vregoff(&p.From))
+		zVRRc(op, uint32(p.To.Reg), uint32(p.Reg), uint32(p.GetFrom3().Reg), m6, m5, m4, asm)
 	}
 }
 
diff --git a/src/cmd/internal/obj/s390x/vector.go b/src/cmd/internal/obj/s390x/vector.go
index e7e36eaf15..966cd04c27 100644
--- a/src/cmd/internal/obj/s390x/vector.go
+++ b/src/cmd/internal/obj/s390x/vector.go
@@ -1027,6 +1027,22 @@ func vop(as obj.As) (opcode, es, cs uint32) {
 		return op_VUPL, 1, 0
 	case AVUPLF:
 		return op_VUPL, 2, 0
+	case AVFMAXDB:
+		return op_VFMAX, 3, 0
+	case AWFMAXDB:
+		return op_VFMAX, 3, 0
+	case AVFMAXSB:
+		return op_VFMAX, 2, 0
+	case AWFMAXSB:
+		return op_VFMAX, 2, 0
+	case AVFMINDB:
+		return op_VFMIN, 3, 0
+	case AWFMINDB:
+		return op_VFMIN, 3, 0
+	case AVFMINSB:
+		return op_VFMIN, 2, 0
+	case AWFMINSB:
+		return op_VFMIN, 2, 0
 	}
 }
 
@@ -1062,7 +1078,11 @@ func singleElementMask(as obj.As) uint32 {
 		AWFSQDB,
 		AWFSDB,
 		AWFTCIDB,
-		AWFIDB:
+		AWFIDB,
+		AWFMAXDB,
+		AWFMAXSB,
+		AWFMINDB,
+		AWFMINSB:
 		return 8
 	}
 	return 0