cmd/internal/obj/s390x: add MULHD instruction

Emulate 64-bit signed high multiplication ((a*b)>>64). To do this
we use the 64-bit unsigned high multiplication method and then
fix the result as shown in Hacker's Delight 2nd ed., chapter 8-3.

Required to enable some division optimizations.

Change-Id: I9194f428e09d3d029cb1afb4715cd5424b5d922e
Reviewed-on: https://go-review.googlesource.com/21774
Reviewed-by: Bill O'Farrell <billotosyr@gmail.com>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>

diff --git a/src/cmd/asm/internal/asm/testdata/s390x.s b/src/cmd/asm/internal/asm/testdata/s390x.s
index 148cd2eaaedb852eb0e4331ddec85bdc37f1f983..f1dc9aff2d9dec233bae90e719f21582314bd068 100644 (file)
--- a/src/cmd/asm/internal/asm/testdata/s390x.s
+++ b/src/cmd/asm/internal/asm/testdata/s390x.s
@@ -61,6 +61,10 @@ TEXT main·foo(SB),7,$16-0 // TEXT main.foo(SB), 7, $16-0
        MULLW   R6, R7, R8            // b9040087b91c0086
        MULLW   $8192, R6             // c26000002000
        MULLW   $8192, R6, R7         // b9040076c27000002000
+       MULHD   R9, R8                // b90400b8b98600a9ebb9003f000ab98000b8b90900abebb8003f000ab98000b9b9e9b08a
+       MULHD   R7, R2, R1            // b90400b2b98600a7ebb7003f000ab98000b2b90900abebb2003f000ab98000b7b9e9b01a
+       MULHDU  R3, R4                // b90400b4b98600a3b904004a
+       MULHDU  R5, R6, R7            // b90400b6b98600a5b904007a
        DIVD    R1, R2                // b90400b2b90d00a1b904002b
        DIVD    R1, R2, R3            // b90400b2b90d00a1b904003b
        DIVW    R4, R5                // b90400b5b91d00a4b904005b

diff --git a/src/cmd/internal/obj/s390x/a.out.go b/src/cmd/internal/obj/s390x/a.out.go
index 2cb03ae603c4e670d66b92e675c6861705cf3ff1..e7256d1d41c56f0f8bdfb7a2763c2acffff23fcd 100644 (file)
--- a/src/cmd/internal/obj/s390x/a.out.go
+++ b/src/cmd/internal/obj/s390x/a.out.go
@@ -218,6 +218,7 @@ const (
        ADIVDU
        AMULLW
        AMULLD
+       AMULHD
        AMULHDU
        ASUB
        ASUBC

diff --git a/src/cmd/internal/obj/s390x/anames.go b/src/cmd/internal/obj/s390x/anames.go
index e79a147a9057360ba57c935540b0aec4f5f4c542..62dd181eda07fa598679a1f11a15cdef16307d2b 100644 (file)
--- a/src/cmd/internal/obj/s390x/anames.go
+++ b/src/cmd/internal/obj/s390x/anames.go
@@ -17,6 +17,7 @@ var Anames = []string{
        "DIVDU",
        "MULLW",
        "MULLD",
+       "MULHD",
        "MULHDU",
        "SUB",
        "SUBC",

diff --git a/src/cmd/internal/obj/s390x/asmz.go b/src/cmd/internal/obj/s390x/asmz.go
index bccd7c3bd89cd9b9081b45d8b7ccab8a36220166..cf3b11424b66730003dcb53c6200ec8c93aea463 100644 (file)
--- a/src/cmd/internal/obj/s390x/asmz.go
+++ b/src/cmd/internal/obj/s390x/asmz.go
@@ -150,6 +150,8 @@ var optab = []Optab{
        Optab{AMULLW, C_REG, C_NONE, C_NONE, C_REG, 2, 0},
        Optab{AMULLW, C_LCON, C_REG, C_NONE, C_REG, 22, 0},
        Optab{AMULLW, C_LCON, C_NONE, C_NONE, C_REG, 22, 0},
+       Optab{AMULHD, C_REG, C_NONE, C_NONE, C_REG, 4, 0},
+       Optab{AMULHD, C_REG, C_REG, C_NONE, C_REG, 4, 0},
        Optab{ASUBC, C_REG, C_REG, C_NONE, C_REG, 10, 0},
        Optab{ASUBC, C_REG, C_NONE, C_NONE, C_REG, 10, 0},
        Optab{ADIVW, C_REG, C_REG, C_NONE, C_REG, 2, 0},
@@ -793,10 +795,11 @@ func buildop(ctxt *obj.Link) {
                case ADIVW:
                        opset(AADDE, r)
                        opset(AMULLD, r)
-                       opset(AMULHDU, r)
                        opset(ADIVD, r)
                        opset(ADIVDU, r)
                        opset(ADIVWU, r)
+               case AMULHD:
+                       opset(AMULHDU, r)
                case AMOVBZ:
                        opset(AMOVH, r)
                        opset(AMOVHZ, r)
@@ -2580,8 +2583,6 @@ func asmout(ctxt *obj.Link, asm *[]byte) {
                        opcode = op_MSGFR
                case AMULLD:
                        opcode = op_MSGR
-               case AMULHDU:
-                       opcode = op_MLGR
                case ADIVW:
                        opcode = op_DSGFR
                case ADIVWU:
@@ -2628,11 +2629,6 @@ func asmout(ctxt *obj.Link, asm *[]byte) {
                        zRRE(opcode, REGTMP, uint32(p.From.Reg), asm)
                        zRRE(op_LGR, uint32(p.To.Reg), REGTMP2, asm)
 
-               case AMULHDU:
-                       zRRE(op_LGR, REGTMP2, uint32(r), asm)
-                       zRRE(opcode, REGTMP, uint32(p.From.Reg), asm)
-                       zRRE(op_LGR, uint32(p.To.Reg), REGTMP, asm)
-
                case AFADD, AFADDS:
                        if r == int(p.To.Reg) {
                                zRRE(opcode, uint32(p.To.Reg), uint32(p.From.Reg), asm)
@@ -2695,6 +2691,28 @@ func asmout(ctxt *obj.Link, asm *[]byte) {
                        zRIL(_a, op_IIHF, uint32(p.To.Reg), uint32(v>>32), asm)
                }
 
+       case 4: // multiply high (a*b)>>64
+               r := p.Reg
+               if r == 0 {
+                       r = p.To.Reg
+               }
+               zRRE(op_LGR, REGTMP2, uint32(r), asm)
+               zRRE(op_MLGR, REGTMP, uint32(p.From.Reg), asm)
+               switch p.As {
+               case AMULHDU:
+                       // Unsigned: move result into correct register.
+                       zRRE(op_LGR, uint32(p.To.Reg), REGTMP, asm)
+               case AMULHD:
+                       // Signed: need to convert result.
+                       // See Hacker's Delight 8-3.
+                       zRSY(op_SRAG, REGTMP2, uint32(p.From.Reg), 0, 63, asm)
+                       zRRE(op_NGR, REGTMP2, uint32(r), asm)
+                       zRRE(op_SGR, REGTMP, REGTMP2, asm)
+                       zRSY(op_SRAG, REGTMP2, uint32(r), 0, 63, asm)
+                       zRRE(op_NGR, REGTMP2, uint32(p.From.Reg), asm)
+                       zRRF(op_SGRK, REGTMP2, 0, uint32(p.To.Reg), REGTMP, asm)
+               }
+
        case 5: // syscall
                zI(op_SVC, 0, asm)