cmd/compile,cmd/asm: on PPC64, generate compares against constant 0

Merge the handling of CMPx r,r,cr and CMPx r,i,cr when assembling.

This prevents generating machine code like cmpd rx,r0 when cmpdi rx,0
is preferred. The preferred form can be fused on Power10 for faster
execution of some instruction sequences.

Likewise, update a common case to use $0 instead of R0 to take
advantage of this.

Change-Id: If2549ca25a5f7d23001885ad444c70d829b3b066
Cq-Include-Trybots: luci.golang.try:gotip-linux-ppc64_power10,gotip-linux-ppc64_power8,gotip-linux-ppc64le_power10,gotip-linux-ppc64le_power8,gotip-linux-ppc64le_power9
Reviewed-on: https://go-review.googlesource.com/c/go/+/585137
Reviewed-by: Dmitri Shuralyov <dmitshur@google.com>
Reviewed-by: Lynn Boger <laboger@linux.vnet.ibm.com>
Reviewed-by: Cherry Mui <cherryyz@google.com>
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com>

3 files changed, 32 insertions, 23 deletions

diff --git a/src/cmd/asm/internal/asm/testdata/ppc64.s b/src/cmd/asm/internal/asm/testdata/ppc64.s
index 8627408f06..7e8c6f9cf2 100644
--- a/src/cmd/asm/internal/asm/testdata/ppc64.s
+++ b/src/cmd/asm/internal/asm/testdata/ppc64.s
@@ -260,13 +260,32 @@ TEXT asmtest(SB),DUPOK|NOSPLIT,$0
 	XORIS $15, R3, R4               // 6c64000f
 	XOR   $983040, R3, R4           // 6c64000f
 
-	// TODO: the order of CR operands don't match
+	// TODO: cleanup inconsistency of printing CMPx opcodes with explicit CR arguments.
 	CMP R3, R4                      // 7c232000
+	CMP R3, R0                      // 7c230000
+	CMP R3, R0, CR1                 // CMP R3,CR1,R0   // 7ca30000
 	CMPU R3, R4                     // 7c232040
+	CMPU R3, R0                     // 7c230040
+	CMPU R3, R0, CR2                // CMPU R3,CR2,R0  // 7d230040
 	CMPW R3, R4                     // 7c032000
+	CMPW R3, R0                     // 7c030000
+	CMPW R3, R0, CR3                // CMPW R3,CR3,R0  // 7d830000
 	CMPWU R3, R4                    // 7c032040
-	CMPB R3,R4,R4                   // 7c6423f8
+	CMPWU R3, R0                    // 7c030040
+	CMPWU R3, R0, CR4               // CMPWU R3,CR4,R0 // 7e030040
+	CMP R3, $0                      // 2c230000
+	CMPU R3, $0                     // 28230000
+	CMPW R3, $0                     // 2c030000
+	CMPWU R3, $0                    // 28030000
+	CMP R3, $0, CR0                 // CMP R3,CR0,$0        // 2c230000
+	CMPU R3, $0, CR1                // CMPU R3,CR1,$0       // 28a30000
+	CMPW R3, $0, CR2                // CMPW R3,CR2,$0       // 2d030000
+	CMPW R3, $-32768, CR2           // CMPW R3,CR2,$-32768  // 2d038000
+	CMPWU R3, $0, CR3               // CMPWU R3,CR3,$0      // 29830000
+	CMPWU R3, $0x8008, CR3          // CMPWU R3,CR3,$32776  // 29838008
+
 	CMPEQB R3,R4,CR6                // 7f0321c0
+	CMPB R3,R4,R4                   // 7c6423f8
 
 	ADD R3, R4                      // 7c841a14
 	ADD R3, R4, R5                  // 7ca41a14
diff --git a/src/cmd/compile/internal/ppc64/ssa.go b/src/cmd/compile/internal/ppc64/ssa.go
index db420b7cb4..d4974ba77e 100644
--- a/src/cmd/compile/internal/ppc64/ssa.go
+++ b/src/cmd/compile/internal/ppc64/ssa.go
@@ -2004,8 +2004,8 @@ func ssaGenBlock(s *ssagen.State, b, next *ssa.Block) {
 		p := s.Prog(ppc64.ACMP)
 		p.From.Type = obj.TYPE_REG
 		p.From.Reg = ppc64.REG_R3
-		p.To.Type = obj.TYPE_REG
-		p.To.Reg = ppc64.REG_R0
+		p.To.Type = obj.TYPE_CONST
+		p.To.Offset = 0
 
 		p = s.Prog(ppc64.ABNE)
 		p.To.Type = obj.TYPE_BRANCH
diff --git a/src/cmd/internal/obj/ppc64/asm9.go b/src/cmd/internal/obj/ppc64/asm9.go
index d9b7c2eed3..74f1772e3d 100644
--- a/src/cmd/internal/obj/ppc64/asm9.go
+++ b/src/cmd/internal/obj/ppc64/asm9.go
@@ -472,12 +472,12 @@ var optabBase = []Optab{
 
 	{as: ACMP, a1: C_REG, a6: C_REG, type_: 70, size: 4},
 	{as: ACMP, a1: C_REG, a2: C_CREG, a6: C_REG, type_: 70, size: 4},
-	{as: ACMP, a1: C_REG, a6: C_S16CON, type_: 71, size: 4},
-	{as: ACMP, a1: C_REG, a2: C_CREG, a6: C_S16CON, type_: 71, size: 4},
+	{as: ACMP, a1: C_REG, a6: C_S16CON, type_: 70, size: 4},
+	{as: ACMP, a1: C_REG, a2: C_CREG, a6: C_S16CON, type_: 70, size: 4},
 	{as: ACMPU, a1: C_REG, a6: C_REG, type_: 70, size: 4},
 	{as: ACMPU, a1: C_REG, a2: C_CREG, a6: C_REG, type_: 70, size: 4},
-	{as: ACMPU, a1: C_REG, a6: C_U16CON, type_: 71, size: 4},
-	{as: ACMPU, a1: C_REG, a2: C_CREG, a6: C_U16CON, type_: 71, size: 4},
+	{as: ACMPU, a1: C_REG, a6: C_U16CON, type_: 70, size: 4},
+	{as: ACMPU, a1: C_REG, a2: C_CREG, a6: C_U16CON, type_: 70, size: 4},
 	{as: AFCMPO, a1: C_FREG, a6: C_FREG, type_: 70, size: 4},
 	{as: AFCMPO, a1: C_FREG, a2: C_CREG, a6: C_FREG, type_: 70, size: 4},
 	{as: ATW, a1: C_32CON, a2: C_REG, a6: C_REG, type_: 60, size: 4},
@@ -3449,23 +3449,13 @@ func asmout(c *ctxt9, p *obj.Prog, o *Optab, out *[5]uint32) {
 
 		o1 = AOP_RRR(OP_MTCRF, uint32(p.From.Reg), 0, 0) | uint32(v)<<12
 
-	case 70: /* [f]cmp r,r,cr*/
-		var r int
-		if p.Reg == 0 {
-			r = 0
-		} else {
-			r = (int(p.Reg) & 7) << 2
-		}
-		o1 = AOP_RRR(c.oprrr(p.As), uint32(r), uint32(p.From.Reg), uint32(p.To.Reg))
-
-	case 71: /* cmp[l] r,i,cr*/
-		var r int
-		if p.Reg == 0 {
-			r = 0
+	case 70: /* cmp* r,r,cr or cmp*i r,i,cr or fcmp f,f,cr or cmpeqb r,r */
+		r := uint32(p.Reg&7) << 2
+		if p.To.Type == obj.TYPE_CONST {
+			o1 = AOP_IRR(c.opirr(p.As), r, uint32(p.From.Reg), uint32(uint16(p.To.Offset)))
 		} else {
-			r = (int(p.Reg) & 7) << 2
+			o1 = AOP_RRR(c.oprrr(p.As), r, uint32(p.From.Reg), uint32(p.To.Reg))
 		}
-		o1 = AOP_RRR(c.opirr(p.As), uint32(r), uint32(p.From.Reg), 0) | uint32(c.regoff(&p.To))&0xffff
 
 	case 72: /* slbmte (Rb+Rs -> slb[Rb]) -> Rs, Rb */
 		o1 = AOP_RRR(c.oprrr(p.As), uint32(p.From.Reg), 0, uint32(p.To.Reg))