From 68819fb6d2bab59e4eadcdf62aa4a2a54417d640 Mon Sep 17 00:00:00 2001
From: Brian Kessler <brian.m.kessler@gmail.com>
Date: Sun, 17 Mar 2019 23:11:00 -0600
Subject: [PATCH] cmd/compile: add signed divisibility by power of 2 rules
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

For powers of two (c=1<<k), the divisibility check x%c == 0 can be made
just by checking the trailing zeroes via a mask x&(c-1)==0 even for signed
integers.  This avoids division fixups when just divisibility check is needed.

To apply this rule the generic divisibility rule for  A%B = A-(A/B*B) is disabled
on the "opt" pass, but this does not affect generated code as this rule is applied
later.

The speed up on amd64 due to elimination of unneccessary fixup code is ~55%:

name                     old time/op  new time/op  delta
DivconstI64-4            2.08ns ± 0%  2.07ns ± 0%     ~     (p=0.079 n=5+5)
DivisiblePow2constI64-4  1.78ns ± 1%  0.81ns ± 1%  -54.55%  (p=0.008 n=5+5)
DivconstU64-4            2.08ns ± 0%  2.08ns ± 0%     ~     (p=1.000 n=5+5)
DivconstI32-4            1.53ns ± 0%  1.53ns ± 0%     ~     (all equal)
DivisiblePow2constI32-4  1.79ns ± 1%  0.81ns ± 4%  -54.75%  (p=0.008 n=5+5)
DivconstU32-4            1.78ns ± 1%  1.78ns ± 1%     ~     (p=1.000 n=5+5)
DivconstI16-4            1.54ns ± 2%  1.53ns ± 0%     ~     (p=0.333 n=5+4)
DivisiblePow2constI16-4  1.78ns ± 0%  0.79ns ± 1%  -55.39%  (p=0.000 n=4+5)
DivconstU16-4            1.00ns ± 5%  0.99ns ± 1%     ~     (p=0.730 n=5+5)
DivconstI8-4             1.54ns ± 0%  1.53ns ± 0%     ~     (p=0.714 n=4+5)
DivisiblePow2constI8-4   1.78ns ± 0%  0.80ns ± 0%  -55.06%  (p=0.000 n=5+4)
DivconstU8-4             0.93ns ± 1%  0.95ns ± 1%   +1.72%  (p=0.024 n=5+5)

A follow-up CL will address the general case of x%c == 0 for signed integers.

Updates #15806

Change-Id: I0d284863774b1bc8c4ce87443bbaec6103e14ef4
Reviewed-on: https://go-review.googlesource.com/c/go/+/168038
Reviewed-by: Keith Randall <khr@golang.org>
---
 .../internal/gc/testdata/arith_test.go        | 291 +++++++++++++++
 .../compile/internal/ssa/gen/generic.rules    |  19 +-
 .../compile/internal/ssa/rewritegeneric.go    | 348 +++++++++++++++++-
 .../compile/internal/test/divconst_test.go    |  26 ++
 test/codegen/arithmetic.go                    |  11 +
 5 files changed, 679 insertions(+), 16 deletions(-)

diff --git a/src/cmd/compile/internal/gc/testdata/arith_test.go b/src/cmd/compile/internal/gc/testdata/arith_test.go
index 728ca56892..1ec9ae02c9 100644
--- a/src/cmd/compile/internal/gc/testdata/arith_test.go
+++ b/src/cmd/compile/internal/gc/testdata/arith_test.go
@@ -7,6 +7,7 @@
 package main
 
 import (
+	"math"
 	"runtime"
 	"testing"
 )
@@ -924,6 +925,7 @@ func TestArithmetic(t *testing.T) {
 	testShiftRemoval(t)
 	testShiftedOps(t)
 	testDivFixUp(t)
+	testDivisibleSignedPow2(t)
 }
 
 // testDivFixUp ensures that signed division fix-ups are being generated.
@@ -952,3 +954,292 @@ func testDivFixUp(t *testing.T) {
 		g64 = z % int64(i)
 	}
 }
+
+//go:noinline
+func divisible_int8_2to1(x int8) bool {
+	return x%(1<<1) == 0
+}
+
+//go:noinline
+func divisible_int8_2to2(x int8) bool {
+	return x%(1<<2) == 0
+}
+
+//go:noinline
+func divisible_int8_2to3(x int8) bool {
+	return x%(1<<3) == 0
+}
+
+//go:noinline
+func divisible_int8_2to4(x int8) bool {
+	return x%(1<<4) == 0
+}
+
+//go:noinline
+func divisible_int8_2to5(x int8) bool {
+	return x%(1<<5) == 0
+}
+
+//go:noinline
+func divisible_int8_2to6(x int8) bool {
+	return x%(1<<6) == 0
+}
+
+//go:noinline
+func divisible_int16_2to1(x int16) bool {
+	return x%(1<<1) == 0
+}
+
+//go:noinline
+func divisible_int16_2to2(x int16) bool {
+	return x%(1<<2) == 0
+}
+
+//go:noinline
+func divisible_int16_2to3(x int16) bool {
+	return x%(1<<3) == 0
+}
+
+//go:noinline
+func divisible_int16_2to4(x int16) bool {
+	return x%(1<<4) == 0
+}
+
+//go:noinline
+func divisible_int16_2to5(x int16) bool {
+	return x%(1<<5) == 0
+}
+
+//go:noinline
+func divisible_int16_2to6(x int16) bool {
+	return x%(1<<6) == 0
+}
+
+//go:noinline
+func divisible_int16_2to7(x int16) bool {
+	return x%(1<<7) == 0
+}
+
+//go:noinline
+func divisible_int16_2to8(x int16) bool {
+	return x%(1<<8) == 0
+}
+
+//go:noinline
+func divisible_int16_2to9(x int16) bool {
+	return x%(1<<9) == 0
+}
+
+//go:noinline
+func divisible_int16_2to10(x int16) bool {
+	return x%(1<<10) == 0
+}
+
+//go:noinline
+func divisible_int16_2to11(x int16) bool {
+	return x%(1<<11) == 0
+}
+
+//go:noinline
+func divisible_int16_2to12(x int16) bool {
+	return x%(1<<12) == 0
+}
+
+//go:noinline
+func divisible_int16_2to13(x int16) bool {
+	return x%(1<<13) == 0
+}
+
+//go:noinline
+func divisible_int16_2to14(x int16) bool {
+	return x%(1<<14) == 0
+}
+
+//go:noinline
+func divisible_int32_2to4(x int32) bool {
+	return x%(1<<4) == 0
+}
+
+//go:noinline
+func divisible_int32_2to15(x int32) bool {
+	return x%(1<<15) == 0
+}
+
+//go:noinline
+func divisible_int32_2to26(x int32) bool {
+	return x%(1<<26) == 0
+}
+
+//go:noinline
+func divisible_int64_2to4(x int64) bool {
+	return x%(1<<4) == 0
+}
+
+//go:noinline
+func divisible_int64_2to15(x int64) bool {
+	return x%(1<<15) == 0
+}
+
+//go:noinline
+func divisible_int64_2to26(x int64) bool {
+	return x%(1<<26) == 0
+}
+
+//go:noinline
+func divisible_int64_2to34(x int64) bool {
+	return x%(1<<34) == 0
+}
+
+//go:noinline
+func divisible_int64_2to48(x int64) bool {
+	return x%(1<<48) == 0
+}
+
+//go:noinline
+func divisible_int64_2to57(x int64) bool {
+	return x%(1<<57) == 0
+}
+
+// testDivisibleSignedPow2 confirms that x%(1<<k)==0 is rewritten correctly
+func testDivisibleSignedPow2(t *testing.T) {
+	var i int64
+	var pow2 = []int64{
+		1,
+		1 << 1,
+		1 << 2,
+		1 << 3,
+		1 << 4,
+		1 << 5,
+		1 << 6,
+		1 << 7,
+		1 << 8,
+		1 << 9,
+		1 << 10,
+		1 << 11,
+		1 << 12,
+		1 << 13,
+		1 << 14,
+	}
+	// exhaustive test for int8
+	for i = math.MinInt8; i <= math.MaxInt8; i++ {
+		if want, got := int8(i)%int8(pow2[1]) == 0, divisible_int8_2to1(int8(i)); got != want {
+			t.Errorf("divisible_int8_2to1(%d) = %v want %v", i, got, want)
+		}
+		if want, got := int8(i)%int8(pow2[2]) == 0, divisible_int8_2to2(int8(i)); got != want {
+			t.Errorf("divisible_int8_2to2(%d) = %v want %v", i, got, want)
+		}
+		if want, got := int8(i)%int8(pow2[3]) == 0, divisible_int8_2to3(int8(i)); got != want {
+			t.Errorf("divisible_int8_2to3(%d) = %v want %v", i, got, want)
+		}
+		if want, got := int8(i)%int8(pow2[4]) == 0, divisible_int8_2to4(int8(i)); got != want {
+			t.Errorf("divisible_int8_2to4(%d) = %v want %v", i, got, want)
+		}
+		if want, got := int8(i)%int8(pow2[5]) == 0, divisible_int8_2to5(int8(i)); got != want {
+			t.Errorf("divisible_int8_2to5(%d) = %v want %v", i, got, want)
+		}
+		if want, got := int8(i)%int8(pow2[6]) == 0, divisible_int8_2to6(int8(i)); got != want {
+			t.Errorf("divisible_int8_2to6(%d) = %v want %v", i, got, want)
+		}
+	}
+	// exhaustive test for int16
+	for i = math.MinInt16; i <= math.MaxInt16; i++ {
+		if want, got := int16(i)%int16(pow2[1]) == 0, divisible_int16_2to1(int16(i)); got != want {
+			t.Errorf("divisible_int16_2to1(%d) = %v want %v", i, got, want)
+		}
+		if want, got := int16(i)%int16(pow2[2]) == 0, divisible_int16_2to2(int16(i)); got != want {
+			t.Errorf("divisible_int16_2to2(%d) = %v want %v", i, got, want)
+		}
+		if want, got := int16(i)%int16(pow2[3]) == 0, divisible_int16_2to3(int16(i)); got != want {
+			t.Errorf("divisible_int16_2to3(%d) = %v want %v", i, got, want)
+		}
+		if want, got := int16(i)%int16(pow2[4]) == 0, divisible_int16_2to4(int16(i)); got != want {
+			t.Errorf("divisible_int16_2to4(%d) = %v want %v", i, got, want)
+		}
+		if want, got := int16(i)%int16(pow2[5]) == 0, divisible_int16_2to5(int16(i)); got != want {
+			t.Errorf("divisible_int16_2to5(%d) = %v want %v", i, got, want)
+		}
+		if want, got := int16(i)%int16(pow2[6]) == 0, divisible_int16_2to6(int16(i)); got != want {
+			t.Errorf("divisible_int16_2to6(%d) = %v want %v", i, got, want)
+		}
+		if want, got := int16(i)%int16(pow2[7]) == 0, divisible_int16_2to7(int16(i)); got != want {
+			t.Errorf("divisible_int16_2to7(%d) = %v want %v", i, got, want)
+		}
+		if want, got := int16(i)%int16(pow2[8]) == 0, divisible_int16_2to8(int16(i)); got != want {
+			t.Errorf("divisible_int16_2to8(%d) = %v want %v", i, got, want)
+		}
+		if want, got := int16(i)%int16(pow2[9]) == 0, divisible_int16_2to9(int16(i)); got != want {
+			t.Errorf("divisible_int16_2to9(%d) = %v want %v", i, got, want)
+		}
+		if want, got := int16(i)%int16(pow2[10]) == 0, divisible_int16_2to10(int16(i)); got != want {
+			t.Errorf("divisible_int16_2to10(%d) = %v want %v", i, got, want)
+		}
+		if want, got := int16(i)%int16(pow2[11]) == 0, divisible_int16_2to11(int16(i)); got != want {
+			t.Errorf("divisible_int16_2to11(%d) = %v want %v", i, got, want)
+		}
+		if want, got := int16(i)%int16(pow2[12]) == 0, divisible_int16_2to12(int16(i)); got != want {
+			t.Errorf("divisible_int16_2to12(%d) = %v want %v", i, got, want)
+		}
+		if want, got := int16(i)%int16(pow2[13]) == 0, divisible_int16_2to13(int16(i)); got != want {
+			t.Errorf("divisible_int16_2to13(%d) = %v want %v", i, got, want)
+		}
+		if want, got := int16(i)%int16(pow2[14]) == 0, divisible_int16_2to14(int16(i)); got != want {
+			t.Errorf("divisible_int16_2to14(%d) = %v want %v", i, got, want)
+		}
+	}
+	// spot check for int32 and int64
+	var (
+		two4  int64 = 1 << 4
+		two15 int64 = 1 << 15
+		two26 int64 = 1 << 26
+		two34 int64 = 1 << 34
+		two48 int64 = 1 << 48
+		two57 int64 = 1 << 57
+	)
+	var xs = []int64{two4, two4 + 3, -3 * two4, -3*two4 + 1,
+		two15, two15 + 3, -3 * two15, -3*two15 + 1,
+		two26, two26 + 37, -5 * two26, -5*two26 + 2,
+		two34, two34 + 356, -7 * two34, -7*two34 + 13,
+		two48, two48 + 3000, -12 * two48, -12*two48 + 1111,
+		two57, two57 + 397654, -15 * two57, -15*two57 + 11234,
+	}
+	for _, x := range xs {
+		if int64(int32(x)) == x {
+			if want, got := int32(x)%int32(two4) == 0, divisible_int32_2to4(int32(x)); got != want {
+				t.Errorf("divisible_int32_2to4(%d) = %v want %v", x, got, want)
+			}
+
+			if want, got := int32(x)%int32(two15) == 0, divisible_int32_2to15(int32(x)); got != want {
+				t.Errorf("divisible_int32_2to15(%d) = %v want %v", x, got, want)
+			}
+
+			if want, got := int32(x)%int32(two26) == 0, divisible_int32_2to26(int32(x)); got != want {
+				t.Errorf("divisible_int32_2to26(%d) = %v want %v", x, got, want)
+			}
+		}
+		// spot check for int64
+		if want, got := x%two4 == 0, divisible_int64_2to4(x); got != want {
+			t.Errorf("divisible_int64_2to4(%d) = %v want %v", x, got, want)
+		}
+
+		if want, got := x%two15 == 0, divisible_int64_2to15(x); got != want {
+			t.Errorf("divisible_int64_2to15(%d) = %v want %v", x, got, want)
+		}
+
+		if want, got := x%two26 == 0, divisible_int64_2to26(x); got != want {
+			t.Errorf("divisible_int64_2to26(%d) = %v want %v", x, got, want)
+		}
+
+		if want, got := x%two34 == 0, divisible_int64_2to34(x); got != want {
+			t.Errorf("divisible_int64_2to34(%d) = %v want %v", x, got, want)
+		}
+
+		if want, got := x%two48 == 0, divisible_int64_2to48(x); got != want {
+			t.Errorf("divisible_int64_2to48(%d) = %v want %v", x, got, want)
+		}
+
+		if want, got := x%two57 == 0, divisible_int64_2to57(x); got != want {
+			t.Errorf("divisible_int64_2to57(%d) = %v want %v", x, got, want)
+		}
+
+	}
+}
diff --git a/src/cmd/compile/internal/ssa/gen/generic.rules b/src/cmd/compile/internal/ssa/gen/generic.rules
index 510cec0f4b..27cb9b62d8 100644
--- a/src/cmd/compile/internal/ssa/gen/generic.rules
+++ b/src/cmd/compile/internal/ssa/gen/generic.rules
@@ -1143,16 +1143,27 @@
 (Mod32 <t> n (Const32 [c])) && c < 0 && c != -1<<31 -> (Mod32 <t> n (Const32 <t> [-c]))
 (Mod64 <t> n (Const64 [c])) && c < 0 && c != -1<<63 -> (Mod64 <t> n (Const64 <t> [-c]))
 
+// Divisibility check for signed integers for power of two constant are simple mask.
+(Eq8 (Mod8 <t> n (Const8 [c])) (Const8 [0])) && n.Op != OpConst8 && isPowerOfTwo(c&0xff)
+  -> (Eq8 (And8 <t> n (Const8 <t> [(c&0xff)-1])) (Const8 <t> [0]))
+(Eq16 (Mod16 <t> n (Const16 [c])) (Const16 [0])) && n.Op != OpConst16 && isPowerOfTwo(c&0xffff)
+  -> (Eq16 (And16 <t> n (Const16 <t> [(c&0xffff)-1])) (Const16 <t> [0]))
+(Eq32 (Mod32 <t> n (Const32 [c])) (Const32 [0])) && n.Op != OpConst32 && isPowerOfTwo(c&0xffffffff)
+  -> (Eq32 (And32 <t> n (Const32 <t> [(c&0xffffffff)-1])) (Const32 <t> [0]))
+(Eq64 (Mod64 <t> n (Const64 [c])) (Const64 [0])) && n.Op != OpConst64 && isPowerOfTwo(c)
+  -> (Eq64 (And64 <t> n (Const64 <t> [c-1])) (Const64 <t> [0]))
+
+
 // All other mods by constants, do A%B = A-(A/B*B).
 // This implements % with two * and a bunch of ancillary ops.
 // One of the * is free if the user's code also computes A/B.
-(Mod8   <t> x (Const8  [c])) && x.Op != OpConst8  && (c > 0 || c == -1<<7)
+(Mod8   <t> x (Const8  [c])) && x.Op != OpConst8  && (c > 0 || c == -1<<7) && v.Block.Func.pass.name != "opt"
   -> (Sub8  x (Mul8  <t> (Div8   <t> x (Const8  <t> [c])) (Const8  <t> [c])))
-(Mod16  <t> x (Const16 [c])) && x.Op != OpConst16 && (c > 0 || c == -1<<15)
+(Mod16  <t> x (Const16 [c])) && x.Op != OpConst16 && (c > 0 || c == -1<<15) && v.Block.Func.pass.name != "opt"
   -> (Sub16 x (Mul16 <t> (Div16  <t> x (Const16 <t> [c])) (Const16 <t> [c])))
-(Mod32  <t> x (Const32 [c])) && x.Op != OpConst32 && (c > 0 || c == -1<<31)
+(Mod32  <t> x (Const32 [c])) && x.Op != OpConst32 && (c > 0 || c == -1<<31) && v.Block.Func.pass.name != "opt"
   -> (Sub32 x (Mul32 <t> (Div32  <t> x (Const32 <t> [c])) (Const32 <t> [c])))
-(Mod64  <t> x (Const64 [c])) && x.Op != OpConst64 && (c > 0 || c == -1<<63)
+(Mod64  <t> x (Const64 [c])) && x.Op != OpConst64 && (c > 0 || c == -1<<63) && v.Block.Func.pass.name != "opt"
   -> (Sub64 x (Mul64 <t> (Div64  <t> x (Const64 <t> [c])) (Const64 <t> [c])))
 (Mod8u  <t> x (Const8  [c])) && x.Op != OpConst8  && c > 0 && umagicOK(8 ,c)
   -> (Sub8  x (Mul8  <t> (Div8u  <t> x (Const8  <t> [c])) (Const8  <t> [c])))
diff --git a/src/cmd/compile/internal/ssa/rewritegeneric.go b/src/cmd/compile/internal/ssa/rewritegeneric.go
index fe2fbb82c0..fd77e6b391 100644
--- a/src/cmd/compile/internal/ssa/rewritegeneric.go
+++ b/src/cmd/compile/internal/ssa/rewritegeneric.go
@@ -98,17 +98,17 @@ func rewriteValuegeneric(v *Value) bool {
 	case OpDiv8u:
 		return rewriteValuegeneric_OpDiv8u_0(v)
 	case OpEq16:
-		return rewriteValuegeneric_OpEq16_0(v)
+		return rewriteValuegeneric_OpEq16_0(v) || rewriteValuegeneric_OpEq16_10(v)
 	case OpEq32:
-		return rewriteValuegeneric_OpEq32_0(v)
+		return rewriteValuegeneric_OpEq32_0(v) || rewriteValuegeneric_OpEq32_10(v)
 	case OpEq32F:
 		return rewriteValuegeneric_OpEq32F_0(v)
 	case OpEq64:
-		return rewriteValuegeneric_OpEq64_0(v)
+		return rewriteValuegeneric_OpEq64_0(v) || rewriteValuegeneric_OpEq64_10(v)
 	case OpEq64F:
 		return rewriteValuegeneric_OpEq64F_0(v)
 	case OpEq8:
-		return rewriteValuegeneric_OpEq8_0(v)
+		return rewriteValuegeneric_OpEq8_0(v) || rewriteValuegeneric_OpEq8_10(v)
 	case OpEqB:
 		return rewriteValuegeneric_OpEqB_0(v)
 	case OpEqInter:
@@ -8951,6 +8951,84 @@ func rewriteValuegeneric_OpEq16_0(v *Value) bool {
 		v.AuxInt = b2i(c == d)
 		return true
 	}
+	// match: (Eq16 (Mod16 <t> n (Const16 [c])) (Const16 [0]))
+	// cond: n.Op != OpConst16 && isPowerOfTwo(c&0xffff)
+	// result: (Eq16 (And16 <t> n (Const16 <t> [(c&0xffff)-1])) (Const16 <t> [0]))
+	for {
+		_ = v.Args[1]
+		v_0 := v.Args[0]
+		if v_0.Op != OpMod16 {
+			break
+		}
+		t := v_0.Type
+		_ = v_0.Args[1]
+		n := v_0.Args[0]
+		v_0_1 := v_0.Args[1]
+		if v_0_1.Op != OpConst16 {
+			break
+		}
+		c := v_0_1.AuxInt
+		v_1 := v.Args[1]
+		if v_1.Op != OpConst16 {
+			break
+		}
+		if v_1.AuxInt != 0 {
+			break
+		}
+		if !(n.Op != OpConst16 && isPowerOfTwo(c&0xffff)) {
+			break
+		}
+		v.reset(OpEq16)
+		v0 := b.NewValue0(v.Pos, OpAnd16, t)
+		v0.AddArg(n)
+		v1 := b.NewValue0(v.Pos, OpConst16, t)
+		v1.AuxInt = (c & 0xffff) - 1
+		v0.AddArg(v1)
+		v.AddArg(v0)
+		v2 := b.NewValue0(v.Pos, OpConst16, t)
+		v2.AuxInt = 0
+		v.AddArg(v2)
+		return true
+	}
+	// match: (Eq16 (Const16 [0]) (Mod16 <t> n (Const16 [c])))
+	// cond: n.Op != OpConst16 && isPowerOfTwo(c&0xffff)
+	// result: (Eq16 (And16 <t> n (Const16 <t> [(c&0xffff)-1])) (Const16 <t> [0]))
+	for {
+		_ = v.Args[1]
+		v_0 := v.Args[0]
+		if v_0.Op != OpConst16 {
+			break
+		}
+		if v_0.AuxInt != 0 {
+			break
+		}
+		v_1 := v.Args[1]
+		if v_1.Op != OpMod16 {
+			break
+		}
+		t := v_1.Type
+		_ = v_1.Args[1]
+		n := v_1.Args[0]
+		v_1_1 := v_1.Args[1]
+		if v_1_1.Op != OpConst16 {
+			break
+		}
+		c := v_1_1.AuxInt
+		if !(n.Op != OpConst16 && isPowerOfTwo(c&0xffff)) {
+			break
+		}
+		v.reset(OpEq16)
+		v0 := b.NewValue0(v.Pos, OpAnd16, t)
+		v0.AddArg(n)
+		v1 := b.NewValue0(v.Pos, OpConst16, t)
+		v1.AuxInt = (c & 0xffff) - 1
+		v0.AddArg(v1)
+		v.AddArg(v0)
+		v2 := b.NewValue0(v.Pos, OpConst16, t)
+		v2.AuxInt = 0
+		v.AddArg(v2)
+		return true
+	}
 	// match: (Eq16 s:(Sub16 x y) (Const16 [0]))
 	// cond: s.Uses == 1
 	// result: (Eq16 x y)
@@ -8977,6 +9055,9 @@ func rewriteValuegeneric_OpEq16_0(v *Value) bool {
 		v.AddArg(y)
 		return true
 	}
+	return false
+}
+func rewriteValuegeneric_OpEq16_10(v *Value) bool {
 	// match: (Eq16 (Const16 [0]) s:(Sub16 x y))
 	// cond: s.Uses == 1
 	// result: (Eq16 x y)
@@ -9183,6 +9264,84 @@ func rewriteValuegeneric_OpEq32_0(v *Value) bool {
 		v.AuxInt = b2i(c == d)
 		return true
 	}
+	// match: (Eq32 (Mod32 <t> n (Const32 [c])) (Const32 [0]))
+	// cond: n.Op != OpConst32 && isPowerOfTwo(c&0xffffffff)
+	// result: (Eq32 (And32 <t> n (Const32 <t> [(c&0xffffffff)-1])) (Const32 <t> [0]))
+	for {
+		_ = v.Args[1]
+		v_0 := v.Args[0]
+		if v_0.Op != OpMod32 {
+			break
+		}
+		t := v_0.Type
+		_ = v_0.Args[1]
+		n := v_0.Args[0]
+		v_0_1 := v_0.Args[1]
+		if v_0_1.Op != OpConst32 {
+			break
+		}
+		c := v_0_1.AuxInt
+		v_1 := v.Args[1]
+		if v_1.Op != OpConst32 {
+			break
+		}
+		if v_1.AuxInt != 0 {
+			break
+		}
+		if !(n.Op != OpConst32 && isPowerOfTwo(c&0xffffffff)) {
+			break
+		}
+		v.reset(OpEq32)
+		v0 := b.NewValue0(v.Pos, OpAnd32, t)
+		v0.AddArg(n)
+		v1 := b.NewValue0(v.Pos, OpConst32, t)
+		v1.AuxInt = (c & 0xffffffff) - 1
+		v0.AddArg(v1)
+		v.AddArg(v0)
+		v2 := b.NewValue0(v.Pos, OpConst32, t)
+		v2.AuxInt = 0
+		v.AddArg(v2)
+		return true
+	}
+	// match: (Eq32 (Const32 [0]) (Mod32 <t> n (Const32 [c])))
+	// cond: n.Op != OpConst32 && isPowerOfTwo(c&0xffffffff)
+	// result: (Eq32 (And32 <t> n (Const32 <t> [(c&0xffffffff)-1])) (Const32 <t> [0]))
+	for {
+		_ = v.Args[1]
+		v_0 := v.Args[0]
+		if v_0.Op != OpConst32 {
+			break
+		}
+		if v_0.AuxInt != 0 {
+			break
+		}
+		v_1 := v.Args[1]
+		if v_1.Op != OpMod32 {
+			break
+		}
+		t := v_1.Type
+		_ = v_1.Args[1]
+		n := v_1.Args[0]
+		v_1_1 := v_1.Args[1]
+		if v_1_1.Op != OpConst32 {
+			break
+		}
+		c := v_1_1.AuxInt
+		if !(n.Op != OpConst32 && isPowerOfTwo(c&0xffffffff)) {
+			break
+		}
+		v.reset(OpEq32)
+		v0 := b.NewValue0(v.Pos, OpAnd32, t)
+		v0.AddArg(n)
+		v1 := b.NewValue0(v.Pos, OpConst32, t)
+		v1.AuxInt = (c & 0xffffffff) - 1
+		v0.AddArg(v1)
+		v.AddArg(v0)
+		v2 := b.NewValue0(v.Pos, OpConst32, t)
+		v2.AuxInt = 0
+		v.AddArg(v2)
+		return true
+	}
 	// match: (Eq32 s:(Sub32 x y) (Const32 [0]))
 	// cond: s.Uses == 1
 	// result: (Eq32 x y)
@@ -9209,6 +9368,9 @@ func rewriteValuegeneric_OpEq32_0(v *Value) bool {
 		v.AddArg(y)
 		return true
 	}
+	return false
+}
+func rewriteValuegeneric_OpEq32_10(v *Value) bool {
 	// match: (Eq32 (Const32 [0]) s:(Sub32 x y))
 	// cond: s.Uses == 1
 	// result: (Eq32 x y)
@@ -9456,6 +9618,84 @@ func rewriteValuegeneric_OpEq64_0(v *Value) bool {
 		v.AuxInt = b2i(c == d)
 		return true
 	}
+	// match: (Eq64 (Mod64 <t> n (Const64 [c])) (Const64 [0]))
+	// cond: n.Op != OpConst64 && isPowerOfTwo(c)
+	// result: (Eq64 (And64 <t> n (Const64 <t> [c-1])) (Const64 <t> [0]))
+	for {
+		_ = v.Args[1]
+		v_0 := v.Args[0]
+		if v_0.Op != OpMod64 {
+			break
+		}
+		t := v_0.Type
+		_ = v_0.Args[1]
+		n := v_0.Args[0]
+		v_0_1 := v_0.Args[1]
+		if v_0_1.Op != OpConst64 {
+			break
+		}
+		c := v_0_1.AuxInt
+		v_1 := v.Args[1]
+		if v_1.Op != OpConst64 {
+			break
+		}
+		if v_1.AuxInt != 0 {
+			break
+		}
+		if !(n.Op != OpConst64 && isPowerOfTwo(c)) {
+			break
+		}
+		v.reset(OpEq64)
+		v0 := b.NewValue0(v.Pos, OpAnd64, t)
+		v0.AddArg(n)
+		v1 := b.NewValue0(v.Pos, OpConst64, t)
+		v1.AuxInt = c - 1
+		v0.AddArg(v1)
+		v.AddArg(v0)
+		v2 := b.NewValue0(v.Pos, OpConst64, t)
+		v2.AuxInt = 0
+		v.AddArg(v2)
+		return true
+	}
+	// match: (Eq64 (Const64 [0]) (Mod64 <t> n (Const64 [c])))
+	// cond: n.Op != OpConst64 && isPowerOfTwo(c)
+	// result: (Eq64 (And64 <t> n (Const64 <t> [c-1])) (Const64 <t> [0]))
+	for {
+		_ = v.Args[1]
+		v_0 := v.Args[0]
+		if v_0.Op != OpConst64 {
+			break
+		}
+		if v_0.AuxInt != 0 {
+			break
+		}
+		v_1 := v.Args[1]
+		if v_1.Op != OpMod64 {
+			break
+		}
+		t := v_1.Type
+		_ = v_1.Args[1]
+		n := v_1.Args[0]
+		v_1_1 := v_1.Args[1]
+		if v_1_1.Op != OpConst64 {
+			break
+		}
+		c := v_1_1.AuxInt
+		if !(n.Op != OpConst64 && isPowerOfTwo(c)) {
+			break
+		}
+		v.reset(OpEq64)
+		v0 := b.NewValue0(v.Pos, OpAnd64, t)
+		v0.AddArg(n)
+		v1 := b.NewValue0(v.Pos, OpConst64, t)
+		v1.AuxInt = c - 1
+		v0.AddArg(v1)
+		v.AddArg(v0)
+		v2 := b.NewValue0(v.Pos, OpConst64, t)
+		v2.AuxInt = 0
+		v.AddArg(v2)
+		return true
+	}
 	// match: (Eq64 s:(Sub64 x y) (Const64 [0]))
 	// cond: s.Uses == 1
 	// result: (Eq64 x y)
@@ -9482,6 +9722,9 @@ func rewriteValuegeneric_OpEq64_0(v *Value) bool {
 		v.AddArg(y)
 		return true
 	}
+	return false
+}
+func rewriteValuegeneric_OpEq64_10(v *Value) bool {
 	// match: (Eq64 (Const64 [0]) s:(Sub64 x y))
 	// cond: s.Uses == 1
 	// result: (Eq64 x y)
@@ -9729,6 +9972,84 @@ func rewriteValuegeneric_OpEq8_0(v *Value) bool {
 		v.AuxInt = b2i(c == d)
 		return true
 	}
+	// match: (Eq8 (Mod8 <t> n (Const8 [c])) (Const8 [0]))
+	// cond: n.Op != OpConst8 && isPowerOfTwo(c&0xff)
+	// result: (Eq8 (And8 <t> n (Const8 <t> [(c&0xff)-1])) (Const8 <t> [0]))
+	for {
+		_ = v.Args[1]
+		v_0 := v.Args[0]
+		if v_0.Op != OpMod8 {
+			break
+		}
+		t := v_0.Type
+		_ = v_0.Args[1]
+		n := v_0.Args[0]
+		v_0_1 := v_0.Args[1]
+		if v_0_1.Op != OpConst8 {
+			break
+		}
+		c := v_0_1.AuxInt
+		v_1 := v.Args[1]
+		if v_1.Op != OpConst8 {
+			break
+		}
+		if v_1.AuxInt != 0 {
+			break
+		}
+		if !(n.Op != OpConst8 && isPowerOfTwo(c&0xff)) {
+			break
+		}
+		v.reset(OpEq8)
+		v0 := b.NewValue0(v.Pos, OpAnd8, t)
+		v0.AddArg(n)
+		v1 := b.NewValue0(v.Pos, OpConst8, t)
+		v1.AuxInt = (c & 0xff) - 1
+		v0.AddArg(v1)
+		v.AddArg(v0)
+		v2 := b.NewValue0(v.Pos, OpConst8, t)
+		v2.AuxInt = 0
+		v.AddArg(v2)
+		return true
+	}
+	// match: (Eq8 (Const8 [0]) (Mod8 <t> n (Const8 [c])))
+	// cond: n.Op != OpConst8 && isPowerOfTwo(c&0xff)
+	// result: (Eq8 (And8 <t> n (Const8 <t> [(c&0xff)-1])) (Const8 <t> [0]))
+	for {
+		_ = v.Args[1]
+		v_0 := v.Args[0]
+		if v_0.Op != OpConst8 {
+			break
+		}
+		if v_0.AuxInt != 0 {
+			break
+		}
+		v_1 := v.Args[1]
+		if v_1.Op != OpMod8 {
+			break
+		}
+		t := v_1.Type
+		_ = v_1.Args[1]
+		n := v_1.Args[0]
+		v_1_1 := v_1.Args[1]
+		if v_1_1.Op != OpConst8 {
+			break
+		}
+		c := v_1_1.AuxInt
+		if !(n.Op != OpConst8 && isPowerOfTwo(c&0xff)) {
+			break
+		}
+		v.reset(OpEq8)
+		v0 := b.NewValue0(v.Pos, OpAnd8, t)
+		v0.AddArg(n)
+		v1 := b.NewValue0(v.Pos, OpConst8, t)
+		v1.AuxInt = (c & 0xff) - 1
+		v0.AddArg(v1)
+		v.AddArg(v0)
+		v2 := b.NewValue0(v.Pos, OpConst8, t)
+		v2.AuxInt = 0
+		v.AddArg(v2)
+		return true
+	}
 	// match: (Eq8 s:(Sub8 x y) (Const8 [0]))
 	// cond: s.Uses == 1
 	// result: (Eq8 x y)
@@ -9755,6 +10076,9 @@ func rewriteValuegeneric_OpEq8_0(v *Value) bool {
 		v.AddArg(y)
 		return true
 	}
+	return false
+}
+func rewriteValuegeneric_OpEq8_10(v *Value) bool {
 	// match: (Eq8 (Const8 [0]) s:(Sub8 x y))
 	// cond: s.Uses == 1
 	// result: (Eq8 x y)
@@ -14908,7 +15232,7 @@ func rewriteValuegeneric_OpMod16_0(v *Value) bool {
 		return true
 	}
 	// match: (Mod16 <t> x (Const16 [c]))
-	// cond: x.Op != OpConst16 && (c > 0 || c == -1<<15)
+	// cond: x.Op != OpConst16 && (c > 0 || c == -1<<15) && v.Block.Func.pass.name != "opt"
 	// result: (Sub16 x (Mul16 <t> (Div16 <t> x (Const16 <t> [c])) (Const16 <t> [c])))
 	for {
 		t := v.Type
@@ -14919,7 +15243,7 @@ func rewriteValuegeneric_OpMod16_0(v *Value) bool {
 			break
 		}
 		c := v_1.AuxInt
-		if !(x.Op != OpConst16 && (c > 0 || c == -1<<15)) {
+		if !(x.Op != OpConst16 && (c > 0 || c == -1<<15) && v.Block.Func.pass.name != "opt") {
 			break
 		}
 		v.reset(OpSub16)
@@ -15087,7 +15411,7 @@ func rewriteValuegeneric_OpMod32_0(v *Value) bool {
 		return true
 	}
 	// match: (Mod32 <t> x (Const32 [c]))
-	// cond: x.Op != OpConst32 && (c > 0 || c == -1<<31)
+	// cond: x.Op != OpConst32 && (c > 0 || c == -1<<31) && v.Block.Func.pass.name != "opt"
 	// result: (Sub32 x (Mul32 <t> (Div32 <t> x (Const32 <t> [c])) (Const32 <t> [c])))
 	for {
 		t := v.Type
@@ -15098,7 +15422,7 @@ func rewriteValuegeneric_OpMod32_0(v *Value) bool {
 			break
 		}
 		c := v_1.AuxInt
-		if !(x.Op != OpConst32 && (c > 0 || c == -1<<31)) {
+		if !(x.Op != OpConst32 && (c > 0 || c == -1<<31) && v.Block.Func.pass.name != "opt") {
 			break
 		}
 		v.reset(OpSub32)
@@ -15287,7 +15611,7 @@ func rewriteValuegeneric_OpMod64_0(v *Value) bool {
 		return true
 	}
 	// match: (Mod64 <t> x (Const64 [c]))
-	// cond: x.Op != OpConst64 && (c > 0 || c == -1<<63)
+	// cond: x.Op != OpConst64 && (c > 0 || c == -1<<63) && v.Block.Func.pass.name != "opt"
 	// result: (Sub64 x (Mul64 <t> (Div64 <t> x (Const64 <t> [c])) (Const64 <t> [c])))
 	for {
 		t := v.Type
@@ -15298,7 +15622,7 @@ func rewriteValuegeneric_OpMod64_0(v *Value) bool {
 			break
 		}
 		c := v_1.AuxInt
-		if !(x.Op != OpConst64 && (c > 0 || c == -1<<63)) {
+		if !(x.Op != OpConst64 && (c > 0 || c == -1<<63) && v.Block.Func.pass.name != "opt") {
 			break
 		}
 		v.reset(OpSub64)
@@ -15487,7 +15811,7 @@ func rewriteValuegeneric_OpMod8_0(v *Value) bool {
 		return true
 	}
 	// match: (Mod8 <t> x (Const8 [c]))
-	// cond: x.Op != OpConst8 && (c > 0 || c == -1<<7)
+	// cond: x.Op != OpConst8 && (c > 0 || c == -1<<7) && v.Block.Func.pass.name != "opt"
 	// result: (Sub8 x (Mul8 <t> (Div8 <t> x (Const8 <t> [c])) (Const8 <t> [c])))
 	for {
 		t := v.Type
@@ -15498,7 +15822,7 @@ func rewriteValuegeneric_OpMod8_0(v *Value) bool {
 			break
 		}
 		c := v_1.AuxInt
-		if !(x.Op != OpConst8 && (c > 0 || c == -1<<7)) {
+		if !(x.Op != OpConst8 && (c > 0 || c == -1<<7) && v.Block.Func.pass.name != "opt") {
 			break
 		}
 		v.reset(OpSub8)
diff --git a/src/cmd/compile/internal/test/divconst_test.go b/src/cmd/compile/internal/test/divconst_test.go
index f585a5b51f..6b0bc4a6fb 100644
--- a/src/cmd/compile/internal/test/divconst_test.go
+++ b/src/cmd/compile/internal/test/divconst_test.go
@@ -8,6 +8,8 @@ import (
 	"testing"
 )
 
+var boolres bool
+
 var i64res int64
 
 func BenchmarkDivconstI64(b *testing.B) {
@@ -16,6 +18,12 @@ func BenchmarkDivconstI64(b *testing.B) {
 	}
 }
 
+func BenchmarkDivisiblePow2constI64(b *testing.B) {
+	for i := 0; i < b.N; i++ {
+		boolres = int64(i)%16 == 0
+	}
+}
+
 var u64res uint64
 
 func BenchmarkDivconstU64(b *testing.B) {
@@ -32,6 +40,12 @@ func BenchmarkDivconstI32(b *testing.B) {
 	}
 }
 
+func BenchmarkDivisiblePow2constI32(b *testing.B) {
+	for i := 0; i < b.N; i++ {
+		boolres = int32(i)%16 == 0
+	}
+}
+
 var u32res uint32
 
 func BenchmarkDivconstU32(b *testing.B) {
@@ -48,6 +62,12 @@ func BenchmarkDivconstI16(b *testing.B) {
 	}
 }
 
+func BenchmarkDivisiblePow2constI16(b *testing.B) {
+	for i := 0; i < b.N; i++ {
+		boolres = int16(i)%16 == 0
+	}
+}
+
 var u16res uint16
 
 func BenchmarkDivconstU16(b *testing.B) {
@@ -64,6 +84,12 @@ func BenchmarkDivconstI8(b *testing.B) {
 	}
 }
 
+func BenchmarkDivisiblePow2constI8(b *testing.B) {
+	for i := 0; i < b.N; i++ {
+		boolres = int8(i)%16 == 0
+	}
+}
+
 var u8res uint8
 
 func BenchmarkDivconstU8(b *testing.B) {
diff --git a/test/codegen/arithmetic.go b/test/codegen/arithmetic.go
index b5976be9d2..535e3349fc 100644
--- a/test/codegen/arithmetic.go
+++ b/test/codegen/arithmetic.go
@@ -185,6 +185,17 @@ func Pow2Mods(n1 uint, n2 int) (uint, int) {
 	return a, b
 }
 
+// Check that signed divisibility checks get converted to AND on low bits
+func Pow2DivisibleSigned(n int) bool {
+	// 386:"TESTL\t[$]63",-"DIVL"
+	// amd64:"TESTQ\t[$]63",-"DIVQ"
+	// arm:"AND\t[$]63",-".*udiv"
+	// arm64:"AND\t[$]63",-"UDIV"
+	// ppc64:"ANDCC\t[$]63"
+	// ppc64le:"ANDCC\t[$]63"
+	return n%64 == 0 // signed
+}
+
 // Check that constant modulo divs get turned into MULs
 func ConstMods(n1 uint, n2 int) (uint, int) {
 	// amd64:"MOVQ\t[$]-1085102592571150095","MULQ",-"DIVQ"