cmd/compile: inline memequal(x, const, sz) for small sizes

This CL adds late expanded memequal(x, const, sz) inlining for 2, 4, 8
bytes size. This PoC is using the same method as CL 248404.
This optimization fires about 100 times in Go compiler (1675 occurrences
reduced to 1574, so -6%).
Also, added unit-tests to codegen/comparisions.go file.

Updates #37275

Change-Id: Ia52808d573cb706d1da8166c5746ede26f46c5da
Reviewed-on: https://go-review.googlesource.com/c/go/+/328291
Reviewed-by: Cherry Mui <[email protected]>
Run-TryBot: Cherry Mui <[email protected]>
Trust: David Chase <[email protected]>

Author: kels-ng

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

@@ -2011,12 +2011,30 @@
 	=> (Invalid)
 
 // for late-expanded calls, recognize memequal applied to a single constant byte
-// TODO figure out breakeven number of bytes for this optimization.
+// Support is limited by 1, 2, 4, 8 byte sizes
 (StaticLECall {callAux} sptr (Addr {scon} (SB)) (Const64 [1]) mem)
   && isSameCall(callAux, "runtime.memequal")
   && symIsRO(scon)
   => (MakeResult (Eq8 (Load <typ.Int8> sptr mem) (Const8 <typ.Int8> [int8(read8(scon,0))])) mem)
 
+(StaticLECall {callAux} sptr (Addr {scon} (SB)) (Const64 [2]) mem)
+  && isSameCall(callAux, "runtime.memequal")
+  && symIsRO(scon)
+  && canLoadUnaligned(config)
+  => (MakeResult (Eq16 (Load <typ.Int16> sptr mem) (Const16 <typ.Int16> [int16(read16(scon,0,config.ctxt.Arch.ByteOrder))])) mem)
+
+(StaticLECall {callAux} sptr (Addr {scon} (SB)) (Const64 [4]) mem)
+  && isSameCall(callAux, "runtime.memequal")
+  && symIsRO(scon)
+  && canLoadUnaligned(config)
+  => (MakeResult (Eq32 (Load <typ.Int32> sptr mem) (Const32 <typ.Int32> [int32(read32(scon,0,config.ctxt.Arch.ByteOrder))])) mem)
+
+(StaticLECall {callAux} sptr (Addr {scon} (SB)) (Const64 [8]) mem)
+  && isSameCall(callAux, "runtime.memequal")
+  && symIsRO(scon)
+  && canLoadUnaligned(config) && config.PtrSize == 8
+  => (MakeResult (Eq64 (Load <typ.Int64> sptr mem) (Const64 <typ.Int64> [int64(read64(scon,0,config.ctxt.Arch.ByteOrder))])) mem)
+
 // Evaluate constant address comparisons.
 (EqPtr  x x) => (ConstBool [true])
 (NeqPtr x x) => (ConstBool [false])

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

@@ -415,6 +415,11 @@ func isSameCall(sym interface{}, name string) bool {
 	return fn != nil && fn.String() == name
 }
 
+// canLoadUnaligned reports if the achitecture supports unaligned load operations
+func canLoadUnaligned(c *Config) bool {
+	return c.ctxt.Arch.Alignment == 1
+}
+
 // nlz returns the number of leading zeros.
 func nlz64(x int64) int { return bits.LeadingZeros64(uint64(x)) }
 func nlz32(x int32) int { return bits.LeadingZeros32(uint32(x)) }

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

@@ -538,3 +538,65 @@ func CmpToOneU_ex2(a uint8, b uint16, c uint32, d uint64) int {
 	}
 	return 0
 }
+
+// Check that small memequals are replaced with eq instructions
+
+func equalConstString1() bool {
+	a := string("A")
+	b := string("Z")
+	// amd64:-".*memequal"
+	// arm64:-".*memequal"
+	return a == b
+}
+
+func equalVarString1(a string) bool {
+	b := string("Z")
+	// amd64:-".*memequal"
+	// arm64:-".*memequal"
+	return a[:1] == b
+}
+
+func equalConstString2() bool {
+	a := string("AA")
+	b := string("ZZ")
+	// amd64:-".*memequal"
+	// arm64:-".*memequal"
+	return a == b
+}
+
+func equalVarString2(a string) bool {
+	b := string("ZZ")
+	// amd64:-".*memequal"
+	// arm64:-".*memequal"
+	return a[:2] == b
+}
+
+func equalConstString4() bool {
+	a := string("AAAA")
+	b := string("ZZZZ")
+	// amd64:-".*memequal"
+	// arm64:-".*memequal"
+	return a == b
+}
+
+func equalVarString4(a string) bool {
+	b := string("ZZZZ")
+	// amd64:-".*memequal"
+	// arm64:-".*memequal"
+	return a[:4] == b
+}
+
+func equalConstString8() bool {
+	a := string("AAAAAAAA")
+	b := string("ZZZZZZZZ")
+	// amd64:-".*memequal"
+	// arm64:-".*memequal"
+	return a == b
+}
+
+func equalVarString8(a string) bool {
+	b := string("ZZZZZZZZ")
+	// amd64:-".*memequal"
+	// arm64:-".*memequal"
+	return a[:8] == b
+}