cmd/compile: combine stores into larger widths

Combine stores into larger widths when it is safe to do so.

Add clobber() function so stray dead uses do not impede the
above rewrites.

Fix bug in loads where all intermediate values depending on
a small load (not just the load itself) must have no other uses.
We really need the small load to be dead after the rewrite..

Fixes #14267

Change-Id: Ib25666cb19777f65082c76238fba51a76beb5d74
Reviewed-on: https://go-review.googlesource.com/22326
Run-TryBot: Keith Randall <[email protected]>
Reviewed-by: Todd Neal <[email protected]>

Author: randall77

src/cmd/compile/internal/gc/testdata/dupLoad.go

@@ -12,7 +12,7 @@ package main
 import "fmt"
 
 //go:noinline
-func read(b []byte) (uint16, uint16) {
+func read1(b []byte) (uint16, uint16) {
 	// There is only a single read of b[0].  The two
 	// returned values must have the same low byte.
 	v := b[0]
@@ -21,7 +21,7 @@ func read(b []byte) (uint16, uint16) {
 
 const N = 100000
 
-func main() {
+func main1() {
 	done := make(chan struct{})
 	b := make([]byte, 2)
 	go func() {
@@ -33,7 +33,7 @@ func main() {
 	}()
 	go func() {
 		for i := 0; i < N; i++ {
-			x, y := read(b)
+			x, y := read1(b)
 			if byte(x) != byte(y) {
 				fmt.Printf("x=%x y=%x\n", x, y)
 				panic("bad")
@@ -44,3 +44,40 @@ func main() {
 	<-done
 	<-done
 }
+
+//go:noinline
+func read2(b []byte) (uint16, uint16) {
+	// There is only a single read of b[1].  The two
+	// returned values must have the same high byte.
+	v := uint16(b[1]) << 8
+	return v, uint16(b[0]) | v
+}
+
+func main2() {
+	done := make(chan struct{})
+	b := make([]byte, 2)
+	go func() {
+		for i := 0; i < N; i++ {
+			b[0] = byte(i)
+			b[1] = byte(i)
+		}
+		done <- struct{}{}
+	}()
+	go func() {
+		for i := 0; i < N; i++ {
+			x, y := read2(b)
+			if x&0xff00 != y&0xff00 {
+				fmt.Printf("x=%x y=%x\n", x, y)
+				panic("bad")
+			}
+		}
+		done <- struct{}{}
+	}()
+	<-done
+	<-done
+}
+
+func main() {
+	main1()
+	main2()
+}