[dev.ssa] cmd/internal/obj/arm64: fix encoding constant into some instructions

When a constant can be encoded in a logical instruction (BITCON), do
it this way instead of using the constant pool. The BITCON testing
code runs faster than table lookup (using map):

(on AMD64 machine, with pseudo random input)
BenchmarkIsBitcon-4   	300000000	         4.04 ns/op
BenchmarkTable-4      	50000000	        27.3 ns/op

The equivalent C code of BITCON testing is formally verified with
model checker CBMC against linear search of the lookup table.

Also handle cases when a constant can be encoded in a MOV instruction.
In this case, materializa the constant into REGTMP without using the
constant pool.

When constants need to be added to the constant pool, make sure to
check whether it fits in 32-bit. If not, store 64-bit.

Both legacy and SSA compiler backends are happy with this.

Fixes #16226.

Change-Id: I883e3069dee093a1cdc40853c42221a198a152b0
Reviewed-on: https://go-review.googlesource.com/26631
Run-TryBot: Cherry Zhang <cherryyz@google.com>
Reviewed-by: David Chase <drchase@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

src/cmd/internal/obj/arm64/a.out.go

@@ -274,6 +274,7 @@ const (
 	C_ADDCON   // 12-bit unsigned, shifted left by 0 or 12
 	C_MOVCON   // generated by a 16-bit constant, optionally inverted and/or shifted by multiple of 16
 	C_BITCON   // bitfield and logical immediate masks
+	C_ABCON0   // could be C_ADDCON0 or C_BITCON
 	C_ABCON    // could be C_ADDCON or C_BITCON
 	C_MBCON    // could be C_MOVCON or C_BITCON
 	C_LCON     // 32-bit constant

src/cmd/internal/obj/arm64/anames7.go

@@ -20,6 +20,7 @@ var cnames7 = []string{
 	"ADDCON",
 	"MOVCON",
 	"BITCON",
+	"ABCON0",
 	"ABCON",
 	"MBCON",
 	"LCON",

src/cmd/internal/obj/arm64/obj7.go

@@ -279,20 +279,30 @@ func progedit(ctxt *obj.Link, p *obj.Prog) {
 	// Rewrite negative immediates as positive immediates with
 	// complementary instruction.
 	switch p.As {
-	case AADD,
-		AADDW,
-		ASUB,
-		ASUBW,
-		ACMP,
-		ACMPW,
-		ACMN,
-		ACMNW:
-		if p.From.Type == obj.NAME_EXTERN && p.From.Offset < 0 {
+	case AADD, ASUB, ACMP, ACMN:
+		if p.From.Type == obj.TYPE_CONST && p.From.Offset < 0 && p.From.Offset != -1<<63 {
 			p.From.Offset = -p.From.Offset
 			p.As = complements[p.As]
 		}
+	case AADDW, ASUBW, ACMPW, ACMNW:
+		if p.From.Type == obj.TYPE_CONST && p.From.Offset < 0 && int32(p.From.Offset) != -1<<31 {
+			p.From.Offset = -p.From.Offset
+			p.As = complements[p.As]
+		}
+	}
 
-		break
+	// For 32-bit logical instruction with constant,
+	// rewrite the high 32-bit to be a repetition of
+	// the low 32-bit, so that the BITCON test can be
+	// shared for both 32-bit and 64-bit. 32-bit ops
+	// will zero the high 32-bit of the destination
+	// register anyway.
+	switch p.As {
+	case AANDW, AORRW, AEORW, AANDSW:
+		if p.From.Type == obj.TYPE_CONST {
+			v := p.From.Offset & 0xffffffff
+			p.From.Offset = v | v<<32
+		}
 	}
 
 	if ctxt.Flag_dynlink {