cmd/compile: fold address calculations into CMPload[const] ops

Makes go binary smaller by 0.2%.

I noticed this in autogenerated equal methods, and there are
probably a lot of those.

Change-Id: I4e04eb3653fbceb9dd6a4eee97ceab1fa4d10b72
Reviewed-on: https://go-review.googlesource.com/135379Reviewed-by: Ilya Tocar <ilya.tocar@intel.com>

src/cmd/compile/internal/ssa/gen/AMD64.rules

@@ -1042,6 +1042,11 @@
 	((ADD|SUB|AND|OR|XOR)Qload [off1+off2] {sym} val base mem)
 ((ADD|SUB|AND|OR|XOR)Lload [off1] {sym} val (ADDQconst [off2] base) mem) && is32Bit(off1+off2) ->
 	((ADD|SUB|AND|OR|XOR)Lload [off1+off2] {sym} val base mem)
+(CMP(Q|L|W|B)load [off1] {sym} (ADDQconst [off2] base) val mem) && is32Bit(off1+off2) ->
+	(CMP(Q|L|W|B)load [off1+off2] {sym} base val mem)
+(CMP(Q|L|W|B)constload [off1] {sym} (ADDQconst [off2] base) mem) && is32Bit(off1+off2) ->
+	(CMP(Q|L|W|B)constload [off1+off2] {sym} base mem)
+
 ((ADD|SUB|MUL|DIV)SSload [off1] {sym} val (ADDQconst [off2] base) mem) && is32Bit(off1+off2) ->
 	((ADD|SUB|MUL|DIV)SSload [off1+off2] {sym} val base mem)
 ((ADD|SUB|MUL|DIV)SDload [off1] {sym} val (ADDQconst [off2] base) mem) && is32Bit(off1+off2) ->
@@ -1088,6 +1093,13 @@
 ((ADD|SUB|AND|OR|XOR)Lload [off1] {sym1} val (LEAQ [off2] {sym2} base) mem)
 	&& is32Bit(off1+off2) && canMergeSym(sym1, sym2) ->
 	((ADD|SUB|AND|OR|XOR)Lload [off1+off2] {mergeSym(sym1,sym2)} val base mem)
+(CMP(Q|L|W|B)load [off1] {sym1} (LEAQ [off2] {sym2} base) val mem)
+	&& is32Bit(off1+off2) && canMergeSym(sym1, sym2) ->
+	(CMP(Q|L|W|B)load [off1+off2] {mergeSym(sym1,sym2)} base val mem)
+(CMP(Q|L|W|B)constload [off1] {sym1} (LEAQ [off2] {sym2} base) mem)
+	&& is32Bit(off1+off2) && canMergeSym(sym1, sym2) ->
+	(CMP(Q|L|W|B)constload [off1+off2] {mergeSym(sym1,sym2)} base mem)
+
 ((ADD|SUB|MUL|DIV)SSload [off1] {sym1} val (LEAQ [off2] {sym2} base) mem)
 	&& is32Bit(off1+off2) && canMergeSym(sym1, sym2) ->
 	((ADD|SUB|MUL|DIV)SSload [off1+off2] {mergeSym(sym1,sym2)} val base mem)

test/codegen/memops.go

+// asmcheck
+
+// Copyright 2018 The Go Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+package codegen
+
+var x [2]bool
+var x8 [2]uint8
+var x16 [2]uint16
+var x32 [2]uint32
+var x64 [2]uint64
+
+func compMem1() int {
+	// amd64:`CMPB\t"".x\+1\(SB\), [$]0`
+	if x[1] {
+		return 1
+	}
+	// amd64:`CMPB\t"".x8\+1\(SB\), [$]7`
+	if x8[1] == 7 {
+		return 1
+	}
+	// amd64:`CMPW\t"".x16\+2\(SB\), [$]7`
+	if x16[1] == 7 {
+		return 1
+	}
+	// amd64:`CMPL\t"".x32\+4\(SB\), [$]7`
+	if x32[1] == 7 {
+		return 1
+	}
+	// amd64:`CMPQ\t"".x64\+8\(SB\), [$]7`
+	if x64[1] == 7 {
+		return 1
+	}
+	return 0
+}
+
+//go:noinline
+func f(x int) bool {
+	return false
+}
+
+//go:noinline
+func f8(x int) int8 {
+	return 0
+}
+
+//go:noinline
+func f16(x int) int16 {
+	return 0
+}
+
+//go:noinline
+func f32(x int) int32 {
+	return 0
+}
+
+//go:noinline
+func f64(x int) int64 {
+	return 0
+}
+
+func compMem2() int {
+	// amd64:`CMPB\t8\(SP\), [$]0`
+	if f(3) {
+		return 1
+	}
+	// amd64:`CMPB\t8\(SP\), [$]7`
+	if f8(3) == 7 {
+		return 1
+	}
+	// amd64:`CMPW\t8\(SP\), [$]7`
+	if f16(3) == 7 {
+		return 1
+	}
+	// amd64:`CMPL\t8\(SP\), [$]7`
+	if f32(3) == 7 {
+		return 1
+	}
+	// amd64:`CMPQ\t8\(SP\), [$]7`
+	if f64(3) == 7 {
+		return 1
+	}
+	return 0
+}