From b60432df143475b575a9da1f2a179ac35c399ed0 Mon Sep 17 00:00:00 2001
From: Wayne Zuo <wdvxdr@golangcn.org>
Date: Wed, 24 Aug 2022 22:17:51 +0800
Subject: [PATCH] cmd/compile: deadcode for LoweredMuluhilo on riscv64

This is a follow up of CL 425101 on RISCV64.

According to RISCV Volume 1, Unprivileged Spec v. 20191213 Chapter 7.1:
If both the high and low bits of the same product are required, then the
recommended code sequence is: MULH[[S]U] rdh, rs1, rs2; MUL rdl, rs1, rs2
(source register specifiers must be in same order and rdh cannot be the
same as rs1 or rs2). Microarchitectures can then fuse these into a single
multiply operation instead of performing two separate multiplies.

So we should not split Muluhilo to separate instructions.

Updates #54607

Change-Id: If47461f3aaaf00e27cd583a9990e144fb8bcdb17
Reviewed-on: https://go-review.googlesource.com/c/go/+/425203
Auto-Submit: Keith Randall <khr@golang.org>
TryBot-Result: Gopher Robot <gobot@golang.org>
Run-TryBot: Wayne Zuo <wdvxdr@golangcn.org>
Reviewed-by: Keith Randall <khr@google.com>
Reviewed-by: Keith Randall <khr@golang.org>
Reviewed-by: Cherry Mui <cherryyz@google.com>
---
 .../compile/internal/ssa/gen/RISCV64.rules    |  4 ++
 .../compile/internal/ssa/rewriteRISCV64.go    | 46 +++++++++++++++++++
 test/codegen/mathbits.go                      |  2 +
 3 files changed, 52 insertions(+)

diff --git a/src/cmd/compile/internal/ssa/gen/RISCV64.rules b/src/cmd/compile/internal/ssa/gen/RISCV64.rules
index dd20be2aeb..5bc47ee1cc 100644
--- a/src/cmd/compile/internal/ssa/gen/RISCV64.rules
+++ b/src/cmd/compile/internal/ssa/gen/RISCV64.rules
@@ -743,6 +743,10 @@
 (SLTI  [x] (MOVDconst [y])) => (MOVDconst [b2i(int64(y) < int64(x))])
 (SLTIU [x] (MOVDconst [y])) => (MOVDconst [b2i(uint64(y) < uint64(x))])
 
+// deadcode for LoweredMuluhilo
+(Select0 m:(LoweredMuluhilo x y)) && m.Uses == 1 => (MULHU x y)
+(Select1 m:(LoweredMuluhilo x y)) && m.Uses == 1 => (MUL x y)
+
 // Merge negation into fused multiply-add and multiply-subtract.
 //
 // Key:
diff --git a/src/cmd/compile/internal/ssa/rewriteRISCV64.go b/src/cmd/compile/internal/ssa/rewriteRISCV64.go
index 2677e99dc0..9253d2d729 100644
--- a/src/cmd/compile/internal/ssa/rewriteRISCV64.go
+++ b/src/cmd/compile/internal/ssa/rewriteRISCV64.go
@@ -605,6 +605,10 @@ func rewriteValueRISCV64(v *Value) bool {
 		return rewriteValueRISCV64_OpRsh8x64(v)
 	case OpRsh8x8:
 		return rewriteValueRISCV64_OpRsh8x8(v)
+	case OpSelect0:
+		return rewriteValueRISCV64_OpSelect0(v)
+	case OpSelect1:
+		return rewriteValueRISCV64_OpSelect1(v)
 	case OpSignExt16to32:
 		v.Op = OpRISCV64MOVHreg
 		return true
@@ -6030,6 +6034,48 @@ func rewriteValueRISCV64_OpRsh8x8(v *Value) bool {
 		return true
 	}
 }
+func rewriteValueRISCV64_OpSelect0(v *Value) bool {
+	v_0 := v.Args[0]
+	// match: (Select0 m:(LoweredMuluhilo x y))
+	// cond: m.Uses == 1
+	// result: (MULHU x y)
+	for {
+		m := v_0
+		if m.Op != OpRISCV64LoweredMuluhilo {
+			break
+		}
+		y := m.Args[1]
+		x := m.Args[0]
+		if !(m.Uses == 1) {
+			break
+		}
+		v.reset(OpRISCV64MULHU)
+		v.AddArg2(x, y)
+		return true
+	}
+	return false
+}
+func rewriteValueRISCV64_OpSelect1(v *Value) bool {
+	v_0 := v.Args[0]
+	// match: (Select1 m:(LoweredMuluhilo x y))
+	// cond: m.Uses == 1
+	// result: (MUL x y)
+	for {
+		m := v_0
+		if m.Op != OpRISCV64LoweredMuluhilo {
+			break
+		}
+		y := m.Args[1]
+		x := m.Args[0]
+		if !(m.Uses == 1) {
+			break
+		}
+		v.reset(OpRISCV64MUL)
+		v.AddArg2(x, y)
+		return true
+	}
+	return false
+}
 func rewriteValueRISCV64_OpSlicemask(v *Value) bool {
 	v_0 := v.Args[0]
 	b := v.Block
diff --git a/test/codegen/mathbits.go b/test/codegen/mathbits.go
index 20c945fbc3..a507d32843 100644
--- a/test/codegen/mathbits.go
+++ b/test/codegen/mathbits.go
@@ -800,12 +800,14 @@ func Mul64(x, y uint64) (hi, lo uint64) {
 
 func Mul64HiOnly(x, y uint64) uint64 {
 	// arm64:"UMULH",-"MUL"
+	// riscv64:"MULHU",-"MUL\t"
 	hi, _ := bits.Mul64(x, y)
 	return hi
 }
 
 func Mul64LoOnly(x, y uint64) uint64 {
 	// arm64:"MUL",-"UMULH"
+	// riscv64:"MUL\t",-"MULHU"
 	_, lo := bits.Mul64(x, y)
 	return lo
 }
-- 
2.50.0