cmd/internal/obj/x86: fix doubled REX byte in POPCNT, others

Tests for this and many other instructions are in a separate followup CL.

For #14068.

Change-Id: I6955315996a34d7fb79369b9d9a0119d11745e85
Reviewed-on: https://go-review.googlesource.com/18849Reviewed-by: Rob Pike <r@golang.org>

src/cmd/internal/obj/x86/asm6.go

@@ -3189,12 +3189,10 @@ func doasm(ctxt *obj.Link, p *obj.Prog) {
 				ctxt.Andptr[0] = Pm
 				ctxt.Andptr = ctxt.Andptr[1:]
 
-			case Pq3: /* 16 bit escape, Rex.w, and opcode escape */
+			case Pq3: /* 16 bit escape and opcode escape + REX.W */
+				ctxt.Rexflag |= Pw
 				ctxt.Andptr[0] = Pe
 				ctxt.Andptr = ctxt.Andptr[1:]
-
-				ctxt.Andptr[0] = Pw
-				ctxt.Andptr = ctxt.Andptr[1:]
 				ctxt.Andptr[0] = Pm
 				ctxt.Andptr = ctxt.Andptr[1:]
 
@@ -3206,12 +3204,10 @@ func doasm(ctxt *obj.Link, p *obj.Prog) {
 				ctxt.Andptr[0] = Pm
 				ctxt.Andptr = ctxt.Andptr[1:]
 
-			case Pfw: /* first escape, Rex.w, and second escape */
+			case Pfw: /* xmm opcode escape + REX.W */
+				ctxt.Rexflag |= Pw
 				ctxt.Andptr[0] = Pf3
 				ctxt.Andptr = ctxt.Andptr[1:]
-
-				ctxt.Andptr[0] = Pw
-				ctxt.Andptr = ctxt.Andptr[1:]
 				ctxt.Andptr[0] = Pm
 				ctxt.Andptr = ctxt.Andptr[1:]