cmd/compile: fix s390x load-combining rules

MOVD{reg,nop} operations (added in CL 36256) inserted to preserve
type information were blocking the load-combining rules. Fix this
by merging type changes into loads wherever possible.

Fixes #19059.

Change-Id: I8a1df06eb0f231b40ae43107d4a3bd0b9c441b59
Reviewed-on: https://go-review.googlesource.com/36843
Run-TryBot: Michael Munday <munday@ca.ibm.com>
Reviewed-by: Keith Randall <khr@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

src/cmd/compile/internal/gc/asm_test.go

@@ -227,7 +227,7 @@ func f(b []byte) uint32 {
 	return binary.LittleEndian.Uint32(b)
 }
 `,
-		[]string{"\tMOVWZ\t\\(.*\\),"},
+		[]string{"\tMOVWBR\t\\(.*\\),"},
 	},
 	{"s390x", "linux", `
 import "encoding/binary"
@@ -235,7 +235,7 @@ func f(b []byte, i int) uint32 {
 	return binary.LittleEndian.Uint32(b[i:])
 }
 `,
-		[]string{"\tMOVWZ\t\\(.*\\)\\(.*\\*1\\),"},
+		[]string{"\tMOVWBR\t\\(.*\\)\\(.*\\*1\\),"},
 	},
 	{"s390x", "linux", `
 import "encoding/binary"
@@ -243,17 +243,48 @@ func f(b []byte) uint64 {
 	return binary.LittleEndian.Uint64(b)
 }
 `,
-		[]string{"\tMOVD\t\\(.*\\),"},
+		[]string{"\tMOVDBR\t\\(.*\\),"},
 	},
 	{"s390x", "linux", `
 import "encoding/binary"
 func f(b []byte, i int) uint64 {
 	return binary.LittleEndian.Uint64(b[i:])
 }
+`,
+		[]string{"\tMOVDBR\t\\(.*\\)\\(.*\\*1\\),"},
+	},
+	{"s390x", "linux", `
+import "encoding/binary"
+func f(b []byte) uint32 {
+	return binary.BigEndian.Uint32(b)
+}
+`,
+		[]string{"\tMOVWZ\t\\(.*\\),"},
+	},
+	{"s390x", "linux", `
+import "encoding/binary"
+func f(b []byte, i int) uint32 {
+	return binary.BigEndian.Uint32(b[i:])
+}
+`,
+		[]string{"\tMOVWZ\t\\(.*\\)\\(.*\\*1\\),"},
+	},
+	{"s390x", "linux", `
+import "encoding/binary"
+func f(b []byte) uint64 {
+	return binary.BigEndian.Uint64(b)
+}
+`,
+		[]string{"\tMOVD\t\\(.*\\),"},
+	},
+	{"s390x", "linux", `
+import "encoding/binary"
+func f(b []byte, i int) uint64 {
+	return binary.BigEndian.Uint64(b[i:])
+}
 `,
 		[]string{"\tMOVD\t\\(.*\\)\\(.*\\*1\\),"},
 	},
-	// TODO: s390x big-endian tests.
 
 	// Structure zeroing.  See issue #18370.
 	{"amd64", "linux", `

src/cmd/compile/internal/ssa/gen/S390X.rules

@@ -457,6 +457,40 @@
 // MOVDnop doesn't emit instruction, only for ensuring the type.
 (MOVDreg x) && x.Uses == 1 -> (MOVDnop x)
 
+// Fold type changes into loads.
+(MOVDreg <t> x:(MOVBZload [off] {sym} ptr mem)) && x.Uses == 1 && clobber(x) -> @x.Block (MOVBZload <t> [off] {sym} ptr mem)
+(MOVDreg <t> x:(MOVBload  [off] {sym} ptr mem)) && x.Uses == 1 && clobber(x) -> @x.Block (MOVBload  <t> [off] {sym} ptr mem)
+(MOVDreg <t> x:(MOVHZload [off] {sym} ptr mem)) && x.Uses == 1 && clobber(x) -> @x.Block (MOVHZload <t> [off] {sym} ptr mem)
+(MOVDreg <t> x:(MOVHload  [off] {sym} ptr mem)) && x.Uses == 1 && clobber(x) -> @x.Block (MOVHload  <t> [off] {sym} ptr mem)
+(MOVDreg <t> x:(MOVWZload [off] {sym} ptr mem)) && x.Uses == 1 && clobber(x) -> @x.Block (MOVWZload <t> [off] {sym} ptr mem)
+(MOVDreg <t> x:(MOVWload  [off] {sym} ptr mem)) && x.Uses == 1 && clobber(x) -> @x.Block (MOVWload  <t> [off] {sym} ptr mem)
+(MOVDreg <t> x:(MOVDload  [off] {sym} ptr mem)) && x.Uses == 1 && clobber(x) -> @x.Block (MOVDload  <t> [off] {sym} ptr mem)
+
+(MOVDnop <t> x:(MOVBZload [off] {sym} ptr mem)) && x.Uses == 1 && clobber(x) -> @x.Block (MOVBZload <t> [off] {sym} ptr mem)
+(MOVDnop <t> x:(MOVBload  [off] {sym} ptr mem)) && x.Uses == 1 && clobber(x) -> @x.Block (MOVBload  <t> [off] {sym} ptr mem)
+(MOVDnop <t> x:(MOVHZload [off] {sym} ptr mem)) && x.Uses == 1 && clobber(x) -> @x.Block (MOVHZload <t> [off] {sym} ptr mem)
+(MOVDnop <t> x:(MOVHload  [off] {sym} ptr mem)) && x.Uses == 1 && clobber(x) -> @x.Block (MOVHload  <t> [off] {sym} ptr mem)
+(MOVDnop <t> x:(MOVWZload [off] {sym} ptr mem)) && x.Uses == 1 && clobber(x) -> @x.Block (MOVWZload <t> [off] {sym} ptr mem)
+(MOVDnop <t> x:(MOVWload  [off] {sym} ptr mem)) && x.Uses == 1 && clobber(x) -> @x.Block (MOVWload  <t> [off] {sym} ptr mem)
+(MOVDnop <t> x:(MOVDload  [off] {sym} ptr mem)) && x.Uses == 1 && clobber(x) -> @x.Block (MOVDload  <t> [off] {sym} ptr mem)
+
+// TODO(mundaym): uncomment rules once signed indexed loads are added.
+(MOVDreg <t> x:(MOVBZloadidx [off] {sym} ptr idx mem)) && x.Uses == 1 && clobber(x) -> @x.Block (MOVBZloadidx <t> [off] {sym} ptr idx mem)
+//(MOVDreg <t> x:(MOVBloadidx  [off] {sym} ptr idx mem)) && x.Uses == 1 && clobber(x) -> @x.Block (MOVBloadidx  <t> [off] {sym} ptr idx mem)
+(MOVDreg <t> x:(MOVHZloadidx [off] {sym} ptr idx mem)) && x.Uses == 1 && clobber(x) -> @x.Block (MOVHZloadidx <t> [off] {sym} ptr idx mem)
+//(MOVDreg <t> x:(MOVHloadidx  [off] {sym} ptr idx mem)) && x.Uses == 1 && clobber(x) -> @x.Block (MOVHloadidx  <t> [off] {sym} ptr idx mem)
+(MOVDreg <t> x:(MOVWZloadidx [off] {sym} ptr idx mem)) && x.Uses == 1 && clobber(x) -> @x.Block (MOVWZloadidx <t> [off] {sym} ptr idx mem)
+//(MOVDreg <t> x:(MOVWloadidx  [off] {sym} ptr idx mem)) && x.Uses == 1 && clobber(x) -> @x.Block (MOVWloadidx  <t> [off] {sym} ptr idx mem)
+(MOVDreg <t> x:(MOVDloadidx  [off] {sym} ptr idx mem)) && x.Uses == 1 && clobber(x) -> @x.Block (MOVDloadidx  <t> [off] {sym} ptr idx mem)
+
+(MOVDnop <t> x:(MOVBZloadidx [off] {sym} ptr idx mem)) && x.Uses == 1 && clobber(x) -> @x.Block (MOVBZloadidx <t> [off] {sym} ptr idx mem)
+//(MOVDnop <t> x:(MOVBloadidx  [off] {sym} ptr idx mem)) && x.Uses == 1 && clobber(x) -> @x.Block (MOVBloadidx  <t> [off] {sym} ptr idx mem)
+(MOVDnop <t> x:(MOVHZloadidx [off] {sym} ptr idx mem)) && x.Uses == 1 && clobber(x) -> @x.Block (MOVHZloadidx <t> [off] {sym} ptr idx mem)
+//(MOVDnop <t> x:(MOVHloadidx  [off] {sym} ptr idx mem)) && x.Uses == 1 && clobber(x) -> @x.Block (MOVHloadidx  <t> [off] {sym} ptr idx mem)
+(MOVDnop <t> x:(MOVWZloadidx [off] {sym} ptr idx mem)) && x.Uses == 1 && clobber(x) -> @x.Block (MOVWZloadidx <t> [off] {sym} ptr idx mem)
+//(MOVDnop <t> x:(MOVWloadidx  [off] {sym} ptr idx mem)) && x.Uses == 1 && clobber(x) -> @x.Block (MOVWloadidx  <t> [off] {sym} ptr idx mem)
+(MOVDnop <t> x:(MOVDloadidx  [off] {sym} ptr idx mem)) && x.Uses == 1 && clobber(x) -> @x.Block (MOVDloadidx  <t> [off] {sym} ptr idx mem)
+
 // Fold sign extensions into conditional moves of constants.
 // Designed to remove the MOVBZreg inserted by the If lowering.
 (MOVBZreg x:(MOVDLT (MOVDconst [c]) (MOVDconst [d]) _)) && int64(uint8(c)) == c && int64(uint8(d)) == d -> (MOVDreg x)