From f7dbbf251980763609a65efe15ef9f8ed0cc5a95 Mon Sep 17 00:00:00 2001
From: Jakub Ciolek <jakub@ciolek.dev>
Date: Sat, 11 Jan 2025 17:56:04 +0100
Subject: [PATCH] cmd/compile: distribute 8 and 16-bit multiplication

Expand the existing rule to cover 8 and 16 bit variants.

compilecmp linux/amd64:

time
time.parseStrictRFC3339.func1 80 -> 70  (-12.50%)
time.Time.appendStrictRFC3339.func1 80 -> 70  (-12.50%)
time.Time.appendStrictRFC3339 439 -> 428  (-2.51%)

time [cmd/compile]
time.parseStrictRFC3339.func1 80 -> 70  (-12.50%)
time.Time.appendStrictRFC3339.func1 80 -> 70  (-12.50%)
time.Time.appendStrictRFC3339 439 -> 428  (-2.51%)

linux/arm64:

time
time.parseStrictRFC3339.func1 changed
time.Time.appendStrictRFC3339.func1 changed
time.Time.appendStrictRFC3339 416 -> 400  (-3.85%)

time [cmd/compile]
time.Time.appendStrictRFC3339 416 -> 400  (-3.85%)
time.parseStrictRFC3339.func1 changed
time.Time.appendStrictRFC3339.func1 changed

Change-Id: I0ad3b2363a9fe8c322dd05fbc13bf151a146d8cb
Reviewed-on: https://go-review.googlesource.com/c/go/+/641756
Auto-Submit: Keith Randall <khr@golang.org>
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com>
Reviewed-by: Cherry Mui <cherryyz@google.com>
Reviewed-by: Keith Randall <khr@google.com>
Reviewed-by: Keith Randall <khr@golang.org>
---
 .../compile/internal/ssa/_gen/generic.rules   |  4 ++
 .../compile/internal/ssa/rewritegeneric.go    | 68 +++++++++++++++++++
 2 files changed, 72 insertions(+)

diff --git a/src/cmd/compile/internal/ssa/_gen/generic.rules b/src/cmd/compile/internal/ssa/_gen/generic.rules
index 9e2e8772c1..8ad246830e 100644
--- a/src/cmd/compile/internal/ssa/_gen/generic.rules
+++ b/src/cmd/compile/internal/ssa/_gen/generic.rules
@@ -353,6 +353,10 @@
   (Add64 (Const64 <t> [c*d]) (Mul64 <t> (Const64 <t> [c]) x))
 (Mul32 (Const32 <t> [c]) (Add32 <t> (Const32 <t> [d]) x)) =>
   (Add32 (Const32 <t> [c*d]) (Mul32 <t> (Const32 <t> [c]) x))
+(Mul16 (Const16 <t> [c]) (Add16 <t> (Const16 <t> [d]) x)) =>
+  (Add16 (Const16 <t> [c*d]) (Mul16 <t> (Const16 <t> [c]) x))
+(Mul8 (Const8 <t> [c]) (Add8 <t> (Const8 <t> [d]) x)) =>
+  (Add8 (Const8 <t> [c*d]) (Mul8 <t> (Const8 <t> [c]) x))
 
 // Rewrite x*y ± x*z  to  x*(y±z)
 (Add(64|32|16|8) <t> (Mul(64|32|16|8) x y) (Mul(64|32|16|8) x z))
diff --git a/src/cmd/compile/internal/ssa/rewritegeneric.go b/src/cmd/compile/internal/ssa/rewritegeneric.go
index 4cb287c9b7..fa771bf27d 100644
--- a/src/cmd/compile/internal/ssa/rewritegeneric.go
+++ b/src/cmd/compile/internal/ssa/rewritegeneric.go
@@ -18194,6 +18194,40 @@ func rewriteValuegeneric_OpMul16(v *Value) bool {
 		}
 		break
 	}
+	// match: (Mul16 (Const16 <t> [c]) (Add16 <t> (Const16 <t> [d]) x))
+	// result: (Add16 (Const16 <t> [c*d]) (Mul16 <t> (Const16 <t> [c]) x))
+	for {
+		for _i0 := 0; _i0 <= 1; _i0, v_0, v_1 = _i0+1, v_1, v_0 {
+			if v_0.Op != OpConst16 {
+				continue
+			}
+			t := v_0.Type
+			c := auxIntToInt16(v_0.AuxInt)
+			if v_1.Op != OpAdd16 || v_1.Type != t {
+				continue
+			}
+			_ = v_1.Args[1]
+			v_1_0 := v_1.Args[0]
+			v_1_1 := v_1.Args[1]
+			for _i1 := 0; _i1 <= 1; _i1, v_1_0, v_1_1 = _i1+1, v_1_1, v_1_0 {
+				if v_1_0.Op != OpConst16 || v_1_0.Type != t {
+					continue
+				}
+				d := auxIntToInt16(v_1_0.AuxInt)
+				x := v_1_1
+				v.reset(OpAdd16)
+				v0 := b.NewValue0(v.Pos, OpConst16, t)
+				v0.AuxInt = int16ToAuxInt(c * d)
+				v1 := b.NewValue0(v.Pos, OpMul16, t)
+				v2 := b.NewValue0(v.Pos, OpConst16, t)
+				v2.AuxInt = int16ToAuxInt(c)
+				v1.AddArg2(v2, x)
+				v.AddArg2(v0, v1)
+				return true
+			}
+		}
+		break
+	}
 	// match: (Mul16 (Const16 [0]) _)
 	// result: (Const16 [0])
 	for {
@@ -18917,6 +18951,40 @@ func rewriteValuegeneric_OpMul8(v *Value) bool {
 		}
 		break
 	}
+	// match: (Mul8 (Const8 <t> [c]) (Add8 <t> (Const8 <t> [d]) x))
+	// result: (Add8 (Const8 <t> [c*d]) (Mul8 <t> (Const8 <t> [c]) x))
+	for {
+		for _i0 := 0; _i0 <= 1; _i0, v_0, v_1 = _i0+1, v_1, v_0 {
+			if v_0.Op != OpConst8 {
+				continue
+			}
+			t := v_0.Type
+			c := auxIntToInt8(v_0.AuxInt)
+			if v_1.Op != OpAdd8 || v_1.Type != t {
+				continue
+			}
+			_ = v_1.Args[1]
+			v_1_0 := v_1.Args[0]
+			v_1_1 := v_1.Args[1]
+			for _i1 := 0; _i1 <= 1; _i1, v_1_0, v_1_1 = _i1+1, v_1_1, v_1_0 {
+				if v_1_0.Op != OpConst8 || v_1_0.Type != t {
+					continue
+				}
+				d := auxIntToInt8(v_1_0.AuxInt)
+				x := v_1_1
+				v.reset(OpAdd8)
+				v0 := b.NewValue0(v.Pos, OpConst8, t)
+				v0.AuxInt = int8ToAuxInt(c * d)
+				v1 := b.NewValue0(v.Pos, OpMul8, t)
+				v2 := b.NewValue0(v.Pos, OpConst8, t)
+				v2.AuxInt = int8ToAuxInt(c)
+				v1.AddArg2(v2, x)
+				v.AddArg2(v0, v1)
+				return true
+			}
+		}
+		break
+	}
 	// match: (Mul8 (Const8 [0]) _)
 	// result: (Const8 [0])
 	for {
-- 
2.51.0