cmd/compile: fix subword store/load elision for MIPS

Apply the fix in CL 44355 to MIPS.

ARM64 has these rules but commented out for performance reason.
Fix the commented rules, in case they are enabled in the future.

Enhance the test so it triggers the failure on ARM and MIPS without
the fix.

Updates #20530.

Change-Id: I82d77448e3939a545fe519d0a29a164f8fa5417c
Reviewed-on: https://go-review.googlesource.com/44430
Run-TryBot: Cherry Zhang <[email protected]>
TryBot-Result: Gobot Gobot <[email protected]>
Reviewed-by: David Chase <[email protected]>

Author: cherrymui

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

@@ -668,14 +668,14 @@
 (MOVWstore [off] {sym} ptr (MOVDconst [0]) mem) -> (MOVWstorezero [off] {sym} ptr mem)
 (MOVDstore [off] {sym} ptr (MOVDconst [0]) mem) -> (MOVDstorezero [off] {sym} ptr mem)
 
-// replace load from same location as preceding store with copy
+// replace load from same location as preceding store with zero/sign extension (or copy in case of full width)
 // these seem to have bad interaction with other rules, resulting in slower code
-//(MOVBload [off] {sym} ptr (MOVBstore [off2] {sym2} ptr2 x _)) && sym == sym2 && off == off2 && isSamePtr(ptr, ptr2) -> x
-//(MOVBUload [off] {sym} ptr (MOVBstore [off2] {sym2} ptr2 x _)) && sym == sym2 && off == off2 && isSamePtr(ptr, ptr2) -> x
-//(MOVHload [off] {sym} ptr (MOVHstore [off2] {sym2} ptr2 x _)) && sym == sym2 && off == off2 && isSamePtr(ptr, ptr2) -> x
-//(MOVHUload [off] {sym} ptr (MOVHstore [off2] {sym2} ptr2 x _)) && sym == sym2 && off == off2 && isSamePtr(ptr, ptr2) -> x
-//(MOVWload [off] {sym} ptr (MOVWstore [off2] {sym2} ptr2 x _)) && sym == sym2 && off == off2 && isSamePtr(ptr, ptr2) -> x
-//(MOVWUload [off] {sym} ptr (MOVWstore [off2] {sym2} ptr2 x _)) && sym == sym2 && off == off2 && isSamePtr(ptr, ptr2) -> x
+//(MOVBload [off] {sym} ptr (MOVBstore [off2] {sym2} ptr2 x _)) && sym == sym2 && off == off2 && isSamePtr(ptr, ptr2) -> (MOVBreg x)
+//(MOVBUload [off] {sym} ptr (MOVBstore [off2] {sym2} ptr2 x _)) && sym == sym2 && off == off2 && isSamePtr(ptr, ptr2) -> (MOVBUreg x)
+//(MOVHload [off] {sym} ptr (MOVHstore [off2] {sym2} ptr2 x _)) && sym == sym2 && off == off2 && isSamePtr(ptr, ptr2) -> (MOVHreg x)
+//(MOVHUload [off] {sym} ptr (MOVHstore [off2] {sym2} ptr2 x _)) && sym == sym2 && off == off2 && isSamePtr(ptr, ptr2) -> (MOVHUreg x)
+//(MOVWload [off] {sym} ptr (MOVWstore [off2] {sym2} ptr2 x _)) && sym == sym2 && off == off2 && isSamePtr(ptr, ptr2) -> (MOVWreg x)
+//(MOVWUload [off] {sym} ptr (MOVWstore [off2] {sym2} ptr2 x _)) && sym == sym2 && off == off2 && isSamePtr(ptr, ptr2) -> (MOVWUreg x)
 //(MOVDload [off] {sym} ptr (MOVDstore [off2] {sym2} ptr2 x _)) && sym == sym2 && off == off2 && isSamePtr(ptr, ptr2) -> x
 //(FMOVSload [off] {sym} ptr (FMOVSstore [off2] {sym2} ptr2 x _)) && sym == sym2 && off == off2 && isSamePtr(ptr, ptr2) -> x
 //(FMOVDload [off] {sym} ptr (FMOVDstore [off2] {sym2} ptr2 x _)) && sym == sym2 && off == off2 && isSamePtr(ptr, ptr2) -> x

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

@@ -524,11 +524,11 @@
 (MOVWstorezero [off1] {sym1} (MOVWaddr [off2] {sym2} ptr) mem) && canMergeSym(sym1,sym2) ->
 	(MOVWstorezero [off1+off2] {mergeSym(sym1,sym2)} ptr mem)
 
-// replace load from same location as preceding store with copy
-(MOVBload [off] {sym} ptr (MOVBstore [off2] {sym2} ptr2 x _)) && sym == sym2 && off == off2 && isSamePtr(ptr, ptr2) && isSigned(x.Type) -> x
-(MOVBUload [off] {sym} ptr (MOVBstore [off2] {sym2} ptr2 x _)) && sym == sym2 && off == off2 && isSamePtr(ptr, ptr2) && !isSigned(x.Type) -> x
-(MOVHload [off] {sym} ptr (MOVHstore [off2] {sym2} ptr2 x _)) && sym == sym2 && off == off2 && isSamePtr(ptr, ptr2) && isSigned(x.Type) -> x
-(MOVHUload [off] {sym} ptr (MOVHstore [off2] {sym2} ptr2 x _)) && sym == sym2 && off == off2 && isSamePtr(ptr, ptr2) && !isSigned(x.Type) -> x
+// replace load from same location as preceding store with zero/sign extension (or copy in case of full width)
+(MOVBload [off] {sym} ptr (MOVBstore [off2] {sym2} ptr2 x _)) && sym == sym2 && off == off2 && isSamePtr(ptr, ptr2) -> (MOVBreg x)
+(MOVBUload [off] {sym} ptr (MOVBstore [off2] {sym2} ptr2 x _)) && sym == sym2 && off == off2 && isSamePtr(ptr, ptr2) -> (MOVBUreg x)
+(MOVHload [off] {sym} ptr (MOVHstore [off2] {sym2} ptr2 x _)) && sym == sym2 && off == off2 && isSamePtr(ptr, ptr2) -> (MOVHreg x)
+(MOVHUload [off] {sym} ptr (MOVHstore [off2] {sym2} ptr2 x _)) && sym == sym2 && off == off2 && isSamePtr(ptr, ptr2) -> (MOVHUreg x)
 (MOVWload [off] {sym} ptr (MOVWstore [off2] {sym2} ptr2 x _)) && sym == sym2 && off == off2 && isSamePtr(ptr, ptr2) -> x
 (MOVFload [off] {sym} ptr (MOVFstore [off2] {sym2} ptr2 x _)) && sym == sym2 && off == off2 && isSamePtr(ptr, ptr2) -> x
 (MOVDload [off] {sym} ptr (MOVDstore [off2] {sym2} ptr2 x _)) && sym == sym2 && off == off2 && isSamePtr(ptr, ptr2) -> x

src/cmd/compile/internal/ssa/rewriteMIPS.go

@@ -8,11 +8,27 @@ package main
 
 var a uint8
 
-func main() {
+//go:noinline
+func f() {
 	b := int8(func() int32 { return -1 }())
 	a = uint8(b)
 	if int32(a) != 255 {
 		// Failing case prints 'got 255 expected 255'
 		println("got", a, "expected 255")
 	}
 }
+
+//go:noinline
+func g() {
+	b := int8(func() uint32 { return 0xffffffff }())
+	a = uint8(b)
+	if int32(a) != 255 {
+		// Failing case prints 'got 255 expected 255'
+		println("got", a, "expected 255")
+	}
+}
+
+func main() {
+	f()
+	g()
+}