cmd/compile/internal/ssa: improve store combine optimization on arm64

Current implementation doesn't consider MOVDreg type operand and fail to combine
it into larger store. This patch fixes the issue.

Fixes #24242

Change-Id: I7d68697f80e76f48c3528ece01a602bf513248ec
Reviewed-on: https://go-review.googlesource.com/98397
Run-TryBot: Giovanni Bajo <rasky@develer.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Cherry Zhang <cherryyz@google.com>

src/cmd/compile/internal/ssa/gen/ARM64.rules

@@ -1520,6 +1520,11 @@
 	&& isSamePtr(ptr0, ptr1)
 	&& clobber(x)
 	-> (MOVHstore [i-1] {s} ptr0 w mem)
+(MOVBstore [i] {s} ptr0 (SRLconst [8] (MOVDreg w)) x:(MOVBstore [i-1] {s} ptr1 w mem))
+	&& x.Uses == 1
+	&& isSamePtr(ptr0, ptr1)
+	&& clobber(x)
+	-> (MOVHstore [i-1] {s} ptr0 w mem)
 (MOVBstore [i] {s} ptr0 (SRLconst [j] w) x:(MOVBstore [i-1] {s} ptr1 w0:(SRLconst [j-8] w) mem))
 	&& x.Uses == 1
 	&& isSamePtr(ptr0, ptr1)
@@ -1530,6 +1535,11 @@
 	&& isSamePtr(ptr0, ptr1)
 	&& clobber(x)
 	-> (MOVHstore [i-1] {s} ptr0 w0 mem)
+(MOVBstore [i] {s} ptr0 (SRLconst [j] (MOVDreg w)) x:(MOVBstore [i-1] {s} ptr1 w0:(SRLconst [j-8] (MOVDreg w)) mem))
+	&& x.Uses == 1
+	&& isSamePtr(ptr0, ptr1)
+	&& clobber(x)
+	-> (MOVHstore [i-1] {s} ptr0 w0 mem)
 (MOVHstore [i] {s} ptr0 (SRLconst [16] w) x:(MOVHstore [i-2] {s} ptr1 w mem))
 	&& x.Uses == 1
 	&& isSamePtr(ptr0, ptr1)
@@ -1540,6 +1550,11 @@
 	&& isSamePtr(ptr0, ptr1)
 	&& clobber(x)
 	-> (MOVWstore [i-2] {s} ptr0 w mem)
+(MOVHstore [i] {s} ptr0 (SRLconst [16] (MOVDreg w)) x:(MOVHstore [i-2] {s} ptr1 w mem))
+	&& x.Uses == 1
+	&& isSamePtr(ptr0, ptr1)
+	&& clobber(x)
+	-> (MOVWstore [i-2] {s} ptr0 w mem)
 (MOVHstore [i] {s} ptr0 (SRLconst [j] w) x:(MOVHstore [i-2] {s} ptr1 w0:(SRLconst [j-16] w) mem))
 	&& x.Uses == 1
 	&& isSamePtr(ptr0, ptr1)
@@ -1589,6 +1604,17 @@
 	&& clobber(x1)
 	&& clobber(x2)
 	-> (MOVWstore [i-3] {s} ptr (REVW <w.Type> w) mem)
+(MOVBstore [i] {s} ptr w
+	x0:(MOVBstore [i-1] {s} ptr (SRLconst [8] (MOVDreg w))
+	x1:(MOVBstore [i-2] {s} ptr (SRLconst [16] (MOVDreg w))
+	x2:(MOVBstore [i-3] {s} ptr (SRLconst [24] (MOVDreg w)) mem))))
+	&& x0.Uses == 1
+	&& x1.Uses == 1
+	&& x2.Uses == 1
+	&& clobber(x0)
+	&& clobber(x1)
+	&& clobber(x2)
+	-> (MOVWstore [i-3] {s} ptr (REVW <w.Type> w) mem)
 (MOVBstore [i] {s} ptr w
 	x0:(MOVBstore [i-1] {s} ptr (SRLconst [8] w)
 	x1:(MOVBstore [i-2] {s} ptr (SRLconst [16] w)
@@ -1608,10 +1634,18 @@
 	&& x.Uses == 1
 	&& clobber(x)
 	-> (MOVHstore [i-1] {s} ptr (REV16W <w.Type> w) mem)
+(MOVBstore [i] {s} ptr w x:(MOVBstore [i-1] {s} ptr (SRLconst [8] (MOVDreg w)) mem))
+	&& x.Uses == 1
+	&& clobber(x)
+	-> (MOVHstore [i-1] {s} ptr (REV16W <w.Type> w) mem)
 (MOVBstore [i] {s} ptr w x:(MOVBstore [i-1] {s} ptr (SRLconst [8] (MOVWUreg w)) mem))
 	&& x.Uses == 1
 	&& clobber(x)
 	-> (MOVHstore [i-1] {s} ptr (REV16W <w.Type> w) mem)
+(MOVBstore [i] {s} ptr w x:(MOVBstore [i-1] {s} ptr (SRLconst [8] (MOVDreg w)) mem))
+	&& x.Uses == 1
+	&& clobber(x)
+	-> (MOVHstore [i-1] {s} ptr (REV16W <w.Type> w) mem)
 
 // FP simplification
 (FNEGS (FMULS x y)) -> (FNMULS x y)

test/codegen/memcombine.go

@@ -110,28 +110,28 @@ func store_le64_idx(b []byte, idx int) {
 
 func store_le32(b []byte) {
 	// amd64:`MOVL\s`
-	// arm64(DISABLED):`MOVW`,-`MOV[BH]`
+	// arm64:`MOVW`,-`MOV[BH]`
 	// ppc64le:`MOVW\s`
 	binary.LittleEndian.PutUint32(b, sink32)
 }
 
 func store_le32_idx(b []byte, idx int) {
 	// amd64:`MOVL\s`
-	// arm64(DISABLED):`MOVW`,-`MOV[BH]`
+	// arm64:`MOVW`,-`MOV[BH]`
 	// ppc64le:`MOVW\s`
 	binary.LittleEndian.PutUint32(b[idx:], sink32)
 }
 
 func store_le16(b []byte) {
 	// amd64:`MOVW\s`
-	// arm64(DISABLED):`MOVH`,-`MOVB`
+	// arm64:`MOVH`,-`MOVB`
 	// ppc64le(DISABLED):`MOVH\s`
 	binary.LittleEndian.PutUint16(b, sink16)
 }
 
 func store_le16_idx(b []byte, idx int) {
 	// amd64:`MOVW\s`
-	// arm64(DISABLED):`MOVH`,-`MOVB`
+	// arm64:`MOVH`,-`MOVB`
 	// ppc64le(DISABLED):`MOVH\s`
 	binary.LittleEndian.PutUint16(b[idx:], sink16)
 }
@@ -150,24 +150,24 @@ func store_be64_idx(b []byte, idx int) {
 
 func store_be32(b []byte) {
 	// amd64:`BSWAPL`,-`SHR.`
-	// arm64(DISABLED):`MOVW`,`REVW`,-`MOV[BH]`
+	// arm64:`MOVW`,`REVW`,-`MOV[BH]`
 	binary.BigEndian.PutUint32(b, sink32)
 }
 
 func store_be32_idx(b []byte, idx int) {
 	// amd64:`BSWAPL`,-`SHR.`
-	// arm64(DISABLED):`MOVW`,`REVW`,-`MOV[BH]`
+	// arm64:`MOVW`,`REVW`,-`MOV[BH]`
 	binary.BigEndian.PutUint32(b[idx:], sink32)
 }
 
 func store_be16(b []byte) {
 	// amd64:`ROLW\s\$8`,-`SHR.`
-	// arm64(DISABLED):`MOVH`,`REV16W`,-`MOVB`
+	// arm64:`MOVH`,`REV16W`,-`MOVB`
 	binary.BigEndian.PutUint16(b, sink16)
 }
 
 func store_be16_idx(b []byte, idx int) {
 	// amd64:`ROLW\s\$8`,-`SHR.`
-	// arm64(DISABLED):`MOVH`,`REV16W`,-`MOVB`
+	// arm64:`MOVH`,`REV16W`,-`MOVB`
 	binary.BigEndian.PutUint16(b[idx:], sink16)
 }