cmd/internal/obj/ppc64: generate smaller machine code for OR/XOR of uint32 values

These binary operations can be done in two sequential instructions instead of loading a
constant into REGTMP and doing the binary op.

Change-Id: Ie0ab863f9e81afad140b92b265bca4d3f0fe90b1
Reviewed-on: https://go-review.googlesource.com/c/go/+/565215
Reviewed-by: Lynn Boger <laboger@linux.vnet.ibm.com>
Reviewed-by: Carlos Amedee <carlos@golang.org>
Reviewed-by: Michael Pratt <mpratt@google.com>
TryBot-Result: Gopher Robot <gobot@golang.org>
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com>
Run-TryBot: Paul Murphy <murp@ibm.com>

diff --git a/src/cmd/asm/internal/asm/testdata/ppc64.s b/src/cmd/asm/internal/asm/testdata/ppc64.s
index 01052b49e7e65de915c90729be60d4fa09266f1f..da0b25c1ac4da8e8d5406deede010c2ae885115b 100644 (file)
--- a/src/cmd/asm/internal/asm/testdata/ppc64.s
+++ b/src/cmd/asm/internal/asm/testdata/ppc64.s
@@ -230,10 +230,10 @@ TEXT asmtest(SB),DUPOK|NOSPLIT,$0
        OR $-32767, R5, R6              // 3be080017fe62b78
        OR $-32768, R6                  // 3be080007fe63378
        OR $-32768, R6, R7              // 3be080007fe73378
-       OR $1234567, R5                 // 641f001263ffd6877fe52b78
-       OR $1234567, R5, R3             // 641f001263ffd6877fe32b78
+       OR $1234567, R5                 // 64a5001260a5d687
+       OR $1234567, R5, R3             // 64a300126063d687
        OR $2147483648, R5, R3          // 64a38000
-       OR $2147483649, R5, R3          // 641f800063ff00017fe32b78
+       OR $2147483649, R5, R3          // 64a3800060630001
        ORIS $255, R3, R4               // 646400ff
        OR $16711680, R3, R4            // 646400ff
 
@@ -249,8 +249,8 @@ TEXT asmtest(SB),DUPOK|NOSPLIT,$0
        XOR $-32767, R5, R6             // 3be080017fe62a78
        XOR $-32768, R6                 // 3be080007fe63278
        XOR $-32768, R6, R7             // 3be080007fe73278
-       XOR $1234567, R5                // 641f001263ffd6877fe52a78
-       XOR $1234567, R5, R3            // 641f001263ffd6877fe32a78
+       XOR $1234567, R5                // 6ca5001268a5d687
+       XOR $1234567, R5, R3            // 6ca300126863d687
        XORIS $15, R3, R4               // 6c64000f
        XOR   $983040, R3, R4           // 6c64000f
 

diff --git a/src/cmd/internal/obj/ppc64/asm9.go b/src/cmd/internal/obj/ppc64/asm9.go
index 189b0fb5a860fdb9cf499cfc6ba7ee93fe83e09e..2793600cd0c582fdff01c095007a44590d2f30fe 100644 (file)
--- a/src/cmd/internal/obj/ppc64/asm9.go
+++ b/src/cmd/internal/obj/ppc64/asm9.go
@@ -156,6 +156,8 @@ var optabBase = []Optab{
        {as: AOR, a1: C_U16CON, a2: C_REG, a6: C_REG, type_: 58, size: 4},
        {as: AOR, a1: C_S16CON, a6: C_REG, type_: 23, size: 8},
        {as: AOR, a1: C_S16CON, a2: C_REG, a6: C_REG, type_: 23, size: 8},
+       {as: AOR, a1: C_U32CON, a2: C_REG, a6: C_REG, type_: 21, size: 8},
+       {as: AOR, a1: C_U32CON, a6: C_REG, type_: 21, size: 8},
        {as: AOR, a1: C_32CON, a6: C_REG, type_: 23, size: 12},
        {as: AOR, a1: C_32CON, a2: C_REG, a6: C_REG, type_: 23, size: 12},
        {as: AORIS, a1: C_U16CON, a6: C_REG, type_: 58, size: 4},
@@ -2284,6 +2286,8 @@ const (
        OP_OR       = 31<<26 | 444<<1 | 0<<10 | 0
        OP_ORI      = 24<<26 | 0<<1 | 0<<10 | 0
        OP_ORIS     = 25<<26 | 0<<1 | 0<<10 | 0
+       OP_XORI     = 26<<26 | 0<<1 | 0<<10 | 0
+       OP_XORIS    = 27<<26 | 0<<1 | 0<<10 | 0
        OP_RLWINM   = 21<<26 | 0<<1 | 0<<10 | 0
        OP_RLWNM    = 23<<26 | 0<<1 | 0<<10 | 0
        OP_SUBF     = 31<<26 | 40<<1 | 0<<10 | 0
@@ -2866,6 +2870,23 @@ func asmout(c *ctxt9, p *obj.Prog, o *Optab, out *[5]uint32) {
                }
                o1 = AOP_IRR(c.opirr(p.As), uint32(p.To.Reg), uint32(r), uint32(v))
 
+       case 21: /* or $u32con,rx[,ry] => oris + ori (similar for xor) */
+               var opu, opl uint32
+               r := uint32(p.Reg)
+               if r == 0 {
+                       r = uint32(p.To.Reg)
+               }
+               switch p.As {
+               case AOR:
+                       opu, opl = OP_ORIS, OP_ORI
+               case AXOR:
+                       opu, opl = OP_XORIS, OP_XORI
+               default:
+                       c.ctxt.Diag("unhandled opcode.\n%v", p)
+               }
+               o1 = LOP_IRR(opu, uint32(p.To.Reg), r, uint32(p.From.Offset>>16))
+               o2 = LOP_IRR(opl, uint32(p.To.Reg), uint32(p.To.Reg), uint32(p.From.Offset)&0xFFFF)
+
        case 22: /* add $lcon/$andcon,r1,r2 ==> oris+ori+add/ori+add, add $s34con,r1 ==> addis+ori+slw+ori+add */
                if p.To.Reg == REGTMP || p.Reg == REGTMP {
                        c.ctxt.Diag("can't synthesize large constant\n%v", p)