cmd/compile: optimize integer-in-range checks

This CL incorporates code from CL 201206 by Josh Bleecher Snyder
(thanks Josh).

This CL restores the integer-in-range optimizations in the SSA
backend. The fuse pass is enhanced to detect inequalities that
could be merged and fuse their associated blocks while the generic
rules optimize them into a single unsigned comparison.

For example, the inequality `x >= 0 && x < 10` will now be optimized
to `unsigned(x) < 10`.

Overall has a fairly positive impact on binary sizes.

name                      old time/op       new time/op       delta
Template                        192ms ± 1%        192ms ± 1%    ~     (p=0.757 n=17+18)
Unicode                        76.6ms ± 2%       76.5ms ± 2%    ~     (p=0.603 n=19+19)
GoTypes                         694ms ± 1%        693ms ± 1%    ~     (p=0.569 n=19+20)
Compiler                        3.26s ± 0%        3.27s ± 0%  +0.25%  (p=0.000 n=20+20)
SSA                             7.41s ± 0%        7.49s ± 0%  +1.10%  (p=0.000 n=17+19)
Flate                           120ms ± 1%        120ms ± 1%  +0.38%  (p=0.003 n=19+19)
GoParser                        152ms ± 1%        152ms ± 1%    ~     (p=0.061 n=17+19)
Reflect                         422ms ± 1%        425ms ± 2%  +0.76%  (p=0.001 n=18+20)
Tar                             167ms ± 1%        167ms ± 0%    ~     (p=0.730 n=18+19)
XML                             233ms ± 4%        231ms ± 1%    ~     (p=0.752 n=20+17)
LinkCompiler                    927ms ± 8%        928ms ± 8%    ~     (p=0.857 n=19+20)
ExternalLinkCompiler            1.81s ± 2%        1.81s ± 2%    ~     (p=0.513 n=19+20)
LinkWithoutDebugCompiler        556ms ±10%        583ms ±13%  +4.95%  (p=0.007 n=20+20)
[Geo mean]                      478ms             481ms       +0.52%

name                      old user-time/op  new user-time/op  delta
Template                        270ms ± 5%        269ms ± 7%    ~     (p=0.925 n=20+20)
Unicode                         134ms ± 7%        131ms ±14%    ~     (p=0.593 n=18+20)
GoTypes                         981ms ± 3%        987ms ± 2%  +0.63%  (p=0.049 n=19+18)
Compiler                        4.50s ± 2%        4.50s ± 1%    ~     (p=0.588 n=19+20)
SSA                             10.6s ± 2%        10.6s ± 1%    ~     (p=0.141 n=20+19)
Flate                           164ms ± 8%        165ms ±10%    ~     (p=0.738 n=20+20)
GoParser                        202ms ± 5%        203ms ± 6%    ~     (p=0.820 n=20+20)
Reflect                         587ms ± 6%        597ms ± 3%    ~     (p=0.087 n=20+18)
Tar                             230ms ± 6%        228ms ± 8%    ~     (p=0.569 n=19+20)
XML                             311ms ± 6%        314ms ± 5%    ~     (p=0.369 n=20+20)
LinkCompiler                    878ms ± 8%        887ms ± 7%    ~     (p=0.289 n=20+20)
ExternalLinkCompiler            1.60s ± 7%        1.60s ± 7%    ~     (p=0.820 n=20+20)
LinkWithoutDebugCompiler        498ms ±12%        489ms ±11%    ~     (p=0.398 n=20+20)
[Geo mean]                      611ms             611ms       +0.05%

name                      old alloc/op      new alloc/op      delta
Template                       36.1MB ± 0%       36.0MB ± 0%  -0.32%  (p=0.000 n=20+20)
Unicode                        28.3MB ± 0%       28.3MB ± 0%  -0.03%  (p=0.000 n=19+20)
GoTypes                         121MB ± 0%        121MB ± 0%    ~     (p=0.226 n=16+20)
Compiler                        563MB ± 0%        563MB ± 0%    ~     (p=0.166 n=20+19)
SSA                            1.32GB ± 0%       1.33GB ± 0%  +0.88%  (p=0.000 n=20+19)
Flate                          22.7MB ± 0%       22.7MB ± 0%  -0.02%  (p=0.033 n=19+20)
GoParser                       27.9MB ± 0%       27.9MB ± 0%  -0.02%  (p=0.001 n=20+20)
Reflect                        78.3MB ± 0%       78.2MB ± 0%  -0.01%  (p=0.019 n=20+20)
Tar                            34.0MB ± 0%       34.0MB ± 0%  -0.04%  (p=0.000 n=20+20)
XML                            43.9MB ± 0%       43.9MB ± 0%  -0.07%  (p=0.000 n=20+19)
LinkCompiler                    205MB ± 0%        205MB ± 0%  +0.44%  (p=0.000 n=20+18)
ExternalLinkCompiler            223MB ± 0%        223MB ± 0%  +0.03%  (p=0.000 n=20+20)
LinkWithoutDebugCompiler        139MB ± 0%        142MB ± 0%  +1.75%  (p=0.000 n=20+20)
[Geo mean]                     93.7MB            93.9MB       +0.20%

name                      old allocs/op     new allocs/op     delta
Template                         363k ± 0%         361k ± 0%  -0.58%  (p=0.000 n=20+19)
Unicode                          329k ± 0%         329k ± 0%  -0.06%  (p=0.000 n=19+20)
GoTypes                         1.28M ± 0%        1.28M ± 0%  -0.01%  (p=0.000 n=20+20)
Compiler                        5.40M ± 0%        5.40M ± 0%  -0.01%  (p=0.000 n=20+20)
SSA                             12.7M ± 0%        12.8M ± 0%  +0.80%  (p=0.000 n=20+20)
Flate                            228k ± 0%         228k ± 0%    ~     (p=0.194 n=20+20)
GoParser                         295k ± 0%         295k ± 0%  -0.04%  (p=0.000 n=20+20)
Reflect                          949k ± 0%         949k ± 0%  -0.01%  (p=0.000 n=20+20)
Tar                              337k ± 0%         337k ± 0%  -0.06%  (p=0.000 n=20+20)
XML                              418k ± 0%         417k ± 0%  -0.17%  (p=0.000 n=20+20)
LinkCompiler                     553k ± 0%         554k ± 0%  +0.22%  (p=0.000 n=20+19)
ExternalLinkCompiler            1.52M ± 0%        1.52M ± 0%  +0.27%  (p=0.000 n=20+20)
LinkWithoutDebugCompiler         186k ± 0%         186k ± 0%  +0.06%  (p=0.000 n=20+20)
[Geo mean]                       723k              723k       +0.03%

name                      old text-bytes    new text-bytes    delta
HelloSize                       828kB ± 0%        828kB ± 0%  -0.01%  (p=0.000 n=20+20)

name                      old data-bytes    new data-bytes    delta
HelloSize                      13.4kB ± 0%       13.4kB ± 0%    ~     (all equal)

name                      old bss-bytes     new bss-bytes     delta
HelloSize                       180kB ± 0%        180kB ± 0%    ~     (all equal)

name                      old exe-bytes     new exe-bytes     delta
HelloSize                      1.23MB ± 0%       1.23MB ± 0%  -0.33%  (p=0.000 n=20+20)

file      before    after     Δ       %
addr2line 4320075   4311883   -8192   -0.190%
asm       5191932   5187836   -4096   -0.079%
buildid   2835338   2831242   -4096   -0.144%
compile   20531717  20569099  +37382  +0.182%
cover     5322511   5318415   -4096   -0.077%
dist      3723749   3719653   -4096   -0.110%
doc       4743515   4739419   -4096   -0.086%
fix       3413960   3409864   -4096   -0.120%
link      6690119   6686023   -4096   -0.061%
nm        4269616   4265520   -4096   -0.096%
pprof     14942189  14929901  -12288  -0.082%
trace     11807164  11790780  -16384  -0.139%
vet       8384104   8388200   +4096   +0.049%
go        15339076  15334980  -4096   -0.027%
total     132258257 132226007 -32250  -0.024%

Fixes #30645.

Change-Id: If551ac5996097f3685870d083151b5843170aab0
Reviewed-on: https://go-review.googlesource.com/c/go/+/165998
Run-TryBot: Michael Munday <mike.munday@ibm.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>

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

@@ -148,7 +148,7 @@ func elimIf(f *Func, loadAddr *sparseSet, dom *Block) bool {
 	// the number of useless instructions executed.
 	const maxfuseinsts = 2
 
-	if len(simple.Values) > maxfuseinsts || !allTrivial(simple) {
+	if len(simple.Values) > maxfuseinsts || !canSpeculativelyExecute(simple) {
 		return false
 	}
 
@@ -305,10 +305,10 @@ func elimIfElse(f *Func, loadAddr *sparseSet, b *Block) bool {
 		return false
 	}
 	yes, no := b.Succs[0].Block(), b.Succs[1].Block()
-	if !isLeafPlain(yes) || len(yes.Values) > 1 || !allTrivial(yes) {
+	if !isLeafPlain(yes) || len(yes.Values) > 1 || !canSpeculativelyExecute(yes) {
 		return false
 	}
-	if !isLeafPlain(no) || len(no.Values) > 1 || !allTrivial(no) {
+	if !isLeafPlain(no) || len(no.Values) > 1 || !canSpeculativelyExecute(no) {
 		return false
 	}
 	if b.Succs[0].Block().Succs[0].Block() != b.Succs[1].Block().Succs[0].Block() {
@@ -415,7 +415,15 @@ func shouldElimIfElse(no, yes, post *Block, arch string) bool {
 	}
 }
 
-func allTrivial(b *Block) bool {
+// canSpeculativelyExecute reports whether every value in the block can
+// be evaluated without causing any observable side effects (memory
+// accesses, panics and so on) except for execution time changes. It
+// also ensures that the block does not contain any phis which we can't
+// speculatively execute.
+// Warning: this function cannot currently detect values that represent
+// instructions the execution of which need to be guarded with CPU
+// hardware feature checks. See issue #34950.
+func canSpeculativelyExecute(b *Block) bool {
 	// don't fuse memory ops, Phi ops, divides (can panic),
 	// or anything else with side-effects
 	for _, v := range b.Values {

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

@@ -428,7 +428,7 @@ var passes = [...]pass{
 	{name: "gcse deadcode", fn: deadcode, required: true}, // clean out after cse and phiopt
 	{name: "nilcheckelim", fn: nilcheckelim},
 	{name: "prove", fn: prove},
-	{name: "fuse plain", fn: fusePlain},
+	{name: "early fuse", fn: fuseEarly},
 	{name: "decompose builtin", fn: decomposeBuiltIn, required: true},
 	{name: "softfloat", fn: softfloat, required: true},
 	{name: "late opt", fn: opt, required: true}, // TODO: split required rules and optimizing rules
@@ -436,7 +436,7 @@ var passes = [...]pass{
 	{name: "generic deadcode", fn: deadcode, required: true}, // remove dead stores, which otherwise mess up store chain
 	{name: "check bce", fn: checkbce},
 	{name: "branchelim", fn: branchelim},
-	{name: "fuse", fn: fuseAll},
+	{name: "late fuse", fn: fuseLate},
 	{name: "dse", fn: dse},
 	{name: "writebarrier", fn: writebarrier, required: true}, // expand write barrier ops
 	{name: "insert resched checks", fn: insertLoopReschedChecks,
@@ -491,7 +491,7 @@ var passOrder = [...]constraint{
 	// allow deadcode to clean up after nilcheckelim
 	{"nilcheckelim", "generic deadcode"},
 	// nilcheckelim generates sequences of plain basic blocks
-	{"nilcheckelim", "fuse"},
+	{"nilcheckelim", "late fuse"},
 	// nilcheckelim relies on opt to rewrite user nil checks
 	{"opt", "nilcheckelim"},
 	// tighten will be most effective when as many values have been removed as possible

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

@@ -8,18 +8,18 @@ import (
 	"cmd/internal/src"
 )
 
-// fusePlain runs fuse(f, fuseTypePlain).
+// fuseEarly runs fuse(f, fuseTypePlain|fuseTypeIntInRange).
-func fusePlain(f *Func) { fuse(f, fuseTypePlain) }
+func fuseEarly(f *Func) { fuse(f, fuseTypePlain|fuseTypeIntInRange) }
 
-// fuseAll runs fuse(f, fuseTypeAll).
+// fuseLate runs fuse(f, fuseTypePlain|fuseTypeIf).
-func fuseAll(f *Func) { fuse(f, fuseTypeAll) }
+func fuseLate(f *Func) { fuse(f, fuseTypePlain|fuseTypeIf) }
 
 type fuseType uint8
 
 const (
 	fuseTypePlain fuseType = 1 << iota
 	fuseTypeIf
-	fuseTypeAll = fuseTypePlain | fuseTypeIf
+	fuseTypeIntInRange
 )
 
 // fuse simplifies control flow by joining basic blocks.
@@ -32,6 +32,9 @@ func fuse(f *Func, typ fuseType) {
 			if typ&fuseTypeIf != 0 {
 				changed = fuseBlockIf(b) || changed
 			}
+			if typ&fuseTypeIntInRange != 0 {
+				changed = fuseIntegerComparisons(b) || changed
+			}
 			if typ&fuseTypePlain != 0 {
 				changed = fuseBlockPlain(b) || changed
 			}

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

@@ -0,0 +1,157 @@
+// Copyright 2019 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 ssa
+
+// fuseIntegerComparisons optimizes inequalities such as '1 <= x && x < 5',
+// which can be optimized to 'unsigned(x-1) < 4'.
+//
+// Look for branch structure like:
+//
+//   p
+//   |\
+//   | b
+//   |/ \
+//   s0 s1
+//
+// In our example, p has control '1 <= x', b has control 'x < 5',
+// and s0 and s1 are the if and else results of the comparison.
+//
+// This will be optimized into:
+//
+//   p
+//    \
+//     b
+//    / \
+//   s0 s1
+//
+// where b has the combined control value 'unsigned(x-1) < 4'.
+// Later passes will then fuse p and b.
+func fuseIntegerComparisons(b *Block) bool {
+	if len(b.Preds) != 1 {
+		return false
+	}
+	p := b.Preds[0].Block()
+	if b.Kind != BlockIf || p.Kind != BlockIf {
+		return false
+	}
+
+	// Don't merge control values if b is likely to be bypassed anyway.
+	if p.Likely == BranchLikely && p.Succs[0].Block() != b {
+		return false
+	}
+	if p.Likely == BranchUnlikely && p.Succs[1].Block() != b {
+		return false
+	}
+
+	// Check if the control values combine to make an integer inequality that
+	// can be further optimized later.
+	bc := b.Controls[0]
+	pc := p.Controls[0]
+	if !areMergeableInequalities(bc, pc) {
+		return false
+	}
+
+	// If the first (true) successors match then we have a disjunction (||).
+	// If the second (false) successors match then we have a conjunction (&&).
+	for i, op := range [2]Op{OpOrB, OpAndB} {
+		if p.Succs[i].Block() != b.Succs[i].Block() {
+			continue
+		}
+
+		// TODO(mundaym): should we also check the cost of executing b?
+		// Currently we might speculatively execute b even if b contains
+		// a lot of instructions. We could just check that len(b.Values)
+		// is lower than a fixed amount. Bear in mind however that the
+		// other optimization passes might yet reduce the cost of b
+		// significantly so we shouldn't be overly conservative.
+		if !canSpeculativelyExecute(b) {
+			return false
+		}
+
+		// Logically combine the control values for p and b.
+		v := b.NewValue0(bc.Pos, op, bc.Type)
+		v.AddArg(pc)
+		v.AddArg(bc)
+
+		// Set the combined control value as the control value for b.
+		b.SetControl(v)
+
+		// Modify p so that it jumps directly to b.
+		p.removeEdge(i)
+		p.Kind = BlockPlain
+		p.Likely = BranchUnknown
+		p.ResetControls()
+
+		return true
+	}
+
+	// TODO: could negate condition(s) to merge controls.
+	return false
+}
+
+// getConstIntArgIndex returns the index of the first argument that is a
+// constant integer or -1 if no such argument exists.
+func getConstIntArgIndex(v *Value) int {
+	for i, a := range v.Args {
+		switch a.Op {
+		case OpConst8, OpConst16, OpConst32, OpConst64:
+			return i
+		}
+	}
+	return -1
+}
+
+// isSignedInequality reports whether op represents the inequality < or ≤
+// in the signed domain.
+func isSignedInequality(v *Value) bool {
+	switch v.Op {
+	case OpLess64, OpLess32, OpLess16, OpLess8,
+		OpLeq64, OpLeq32, OpLeq16, OpLeq8:
+		return true
+	}
+	return false
+}
+
+// isUnsignedInequality reports whether op represents the inequality < or ≤
+// in the unsigned domain.
+func isUnsignedInequality(v *Value) bool {
+	switch v.Op {
+	case OpLess64U, OpLess32U, OpLess16U, OpLess8U,
+		OpLeq64U, OpLeq32U, OpLeq16U, OpLeq8U:
+		return true
+	}
+	return false
+}
+
+func areMergeableInequalities(x, y *Value) bool {
+	// We need both inequalities to be either in the signed or unsigned domain.
+	// TODO(mundaym): it would also be good to merge when we have an Eq op that
+	// could be transformed into a Less/Leq. For example in the unsigned
+	// domain 'x == 0 || 3 < x' is equivalent to 'x <= 0 || 3 < x'
+	inequalityChecks := [...]func(*Value) bool{
+		isSignedInequality,
+		isUnsignedInequality,
+	}
+	for _, f := range inequalityChecks {
+		if !f(x) || !f(y) {
+			continue
+		}
+
+		// Check that both inequalities are comparisons with constants.
+		xi := getConstIntArgIndex(x)
+		if xi < 0 {
+			return false
+		}
+		yi := getConstIntArgIndex(y)
+		if yi < 0 {
+			return false
+		}
+
+		// Check that the non-constant arguments to the inequalities
+		// are the same.
+		return x.Args[xi^1] == y.Args[yi^1]
+	}
+	return false
+}

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

@@ -26,7 +26,7 @@ func TestFuseEliminatesOneBranch(t *testing.T) {
 			Exit("mem")))
 
 	CheckFunc(fun.f)
-	fuseAll(fun.f)
+	fuseLate(fun.f)
 
 	for _, b := range fun.f.Blocks {
 		if b == fun.blocks["then"] && b.Kind != BlockInvalid {
@@ -56,7 +56,7 @@ func TestFuseEliminatesBothBranches(t *testing.T) {
 			Exit("mem")))
 
 	CheckFunc(fun.f)
-	fuseAll(fun.f)
+	fuseLate(fun.f)
 
 	for _, b := range fun.f.Blocks {
 		if b == fun.blocks["then"] && b.Kind != BlockInvalid {
@@ -90,7 +90,7 @@ func TestFuseHandlesPhis(t *testing.T) {
 			Exit("mem")))
 
 	CheckFunc(fun.f)
-	fuseAll(fun.f)
+	fuseLate(fun.f)
 
 	for _, b := range fun.f.Blocks {
 		if b == fun.blocks["then"] && b.Kind != BlockInvalid {
@@ -122,7 +122,7 @@ func TestFuseEliminatesEmptyBlocks(t *testing.T) {
 		))
 
 	CheckFunc(fun.f)
-	fuseAll(fun.f)
+	fuseLate(fun.f)
 
 	for k, b := range fun.blocks {
 		if k[:1] == "z" && b.Kind != BlockInvalid {
@@ -153,7 +153,7 @@ func TestFuseSideEffects(t *testing.T) {
 			Goto("loop")))
 
 	CheckFunc(fun.f)
-	fuseAll(fun.f)
+	fuseLate(fun.f)
 
 	for _, b := range fun.f.Blocks {
 		if b == fun.blocks["then"] && b.Kind == BlockInvalid {
@@ -196,7 +196,7 @@ func BenchmarkFuse(b *testing.B) {
 			b.ResetTimer()
 			for i := 0; i < b.N; i++ {
 				fun := c.Fun("entry", blocks...)
-				fuseAll(fun.f)
+				fuseLate(fun.f)
 			}
 		})
 	}

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

@@ -254,6 +254,54 @@
 (Neq16 (Const16 <t> [c]) (Add16 (Const16 <t> [d]) x)) -> (Neq16 (Const16 <t> [int64(int16(c-d))]) x)
 (Neq8  (Const8  <t> [c]) (Add8  (Const8  <t> [d]) x)) -> (Neq8 (Const8 <t> [int64(int8(c-d))]) x)
 
+// signed integer range: ( c <= x && x (<|<=) d ) -> ( unsigned(x-c) (<|<=) unsigned(d-c) )
+(AndB (Leq64 (Const64 [c]) x) ((Less|Leq)64 x (Const64 [d]))) && d >= c -> ((Less|Leq)64U (Sub64 <x.Type> x (Const64 <x.Type> [c])) (Const64 <x.Type> [d-c]))
+(AndB (Leq32 (Const32 [c]) x) ((Less|Leq)32 x (Const32 [d]))) && d >= c -> ((Less|Leq)32U (Sub32 <x.Type> x (Const32 <x.Type> [c])) (Const32 <x.Type> [d-c]))
+(AndB (Leq16 (Const16 [c]) x) ((Less|Leq)16 x (Const16 [d]))) && d >= c -> ((Less|Leq)16U (Sub16 <x.Type> x (Const16 <x.Type> [c])) (Const16 <x.Type> [d-c]))
+(AndB (Leq8  (Const8  [c]) x) ((Less|Leq)8  x (Const8  [d]))) && d >= c -> ((Less|Leq)8U  (Sub8  <x.Type> x (Const8  <x.Type> [c])) (Const8  <x.Type> [d-c]))
+
+// signed integer range: ( c < x && x (<|<=) d ) -> ( unsigned(x-(c+1)) (<|<=) unsigned(d-(c+1)) )
+(AndB (Less64 (Const64 [c]) x) ((Less|Leq)64 x (Const64 [d]))) && d >= c+1 && int64(c+1) > int64(c) -> ((Less|Leq)64U (Sub64 <x.Type> x (Const64 <x.Type> [c+1])) (Const64 <x.Type> [d-c-1]))
+(AndB (Less32 (Const32 [c]) x) ((Less|Leq)32 x (Const32 [d]))) && d >= c+1 && int32(c+1) > int32(c) -> ((Less|Leq)32U (Sub32 <x.Type> x (Const32 <x.Type> [c+1])) (Const32 <x.Type> [d-c-1]))
+(AndB (Less16 (Const16 [c]) x) ((Less|Leq)16 x (Const16 [d]))) && d >= c+1 && int16(c+1) > int16(c) -> ((Less|Leq)16U (Sub16 <x.Type> x (Const16 <x.Type> [c+1])) (Const16 <x.Type> [d-c-1]))
+(AndB (Less8  (Const8  [c]) x) ((Less|Leq)8  x (Const8  [d]))) && d >= c+1 && int8(c+1)  > int8(c)  -> ((Less|Leq)8U  (Sub8  <x.Type> x (Const8  <x.Type> [c+1])) (Const8  <x.Type> [d-c-1]))
+
+// unsigned integer range: ( c <= x && x (<|<=) d ) -> ( x-c (<|<=) d-c )
+(AndB (Leq64U (Const64 [c]) x) ((Less|Leq)64U x (Const64 [d]))) && uint64(d) >= uint64(c) -> ((Less|Leq)64U (Sub64 <x.Type> x (Const64 <x.Type> [c])) (Const64 <x.Type> [d-c]))
+(AndB (Leq32U (Const32 [c]) x) ((Less|Leq)32U x (Const32 [d]))) && uint32(d) >= uint32(c) -> ((Less|Leq)32U (Sub32 <x.Type> x (Const32 <x.Type> [c])) (Const32 <x.Type> [int64(int32(d-c))]))
+(AndB (Leq16U (Const16 [c]) x) ((Less|Leq)16U x (Const16 [d]))) && uint16(d) >= uint16(c) -> ((Less|Leq)16U (Sub16 <x.Type> x (Const16 <x.Type> [c])) (Const16 <x.Type> [int64(int16(d-c))]))
+(AndB (Leq8U  (Const8  [c]) x) ((Less|Leq)8U  x (Const8  [d]))) && uint8(d)  >= uint8(c)  -> ((Less|Leq)8U  (Sub8  <x.Type> x (Const8  <x.Type> [c])) (Const8  <x.Type> [int64(int8(d-c))]))
+
+// unsigned integer range: ( c < x && x (<|<=) d ) -> ( x-(c+1) (<|<=) d-(c+1) )
+(AndB (Less64U (Const64 [c]) x) ((Less|Leq)64U x (Const64 [d]))) && uint64(d) >= uint64(c+1) && uint64(c+1) > uint64(c) -> ((Less|Leq)64U (Sub64 <x.Type> x (Const64 <x.Type> [c+1])) (Const64 <x.Type> [d-c-1]))
+(AndB (Less32U (Const32 [c]) x) ((Less|Leq)32U x (Const32 [d]))) && uint32(d) >= uint32(c+1) && uint32(c+1) > uint32(c) -> ((Less|Leq)32U (Sub32 <x.Type> x (Const32 <x.Type> [int64(int32(c+1))])) (Const32 <x.Type> [int64(int32(d-c-1))]))
+(AndB (Less16U (Const16 [c]) x) ((Less|Leq)16U x (Const16 [d]))) && uint16(d) >= uint16(c+1) && uint16(c+1) > uint16(c) -> ((Less|Leq)16U (Sub16 <x.Type> x (Const16 <x.Type> [int64(int16(c+1))])) (Const16 <x.Type> [int64(int16(d-c-1))]))
+(AndB (Less8U  (Const8  [c]) x) ((Less|Leq)8U  x (Const8  [d]))) && uint8(d)  >= uint8(c+1)  && uint8(c+1)  > uint8(c)  -> ((Less|Leq)8U  (Sub8  <x.Type> x (Const8  <x.Type> [int64(int8(c+1))]))  (Const8  <x.Type> [int64(int8(d-c-1))]))
+
+// signed integer range: ( c (<|<=) x || x < d ) -> ( unsigned(c-d) (<|<=) unsigned(x-d) )
+(OrB ((Less|Leq)64 (Const64 [c]) x) (Less64 x (Const64 [d]))) && c >= d -> ((Less|Leq)64U (Const64 <x.Type> [c-d]) (Sub64 <x.Type> x (Const64 <x.Type> [d])))
+(OrB ((Less|Leq)32 (Const32 [c]) x) (Less32 x (Const32 [d]))) && c >= d -> ((Less|Leq)32U (Const32 <x.Type> [c-d]) (Sub32 <x.Type> x (Const32 <x.Type> [d])))
+(OrB ((Less|Leq)16 (Const16 [c]) x) (Less16 x (Const16 [d]))) && c >= d -> ((Less|Leq)16U (Const16 <x.Type> [c-d]) (Sub16 <x.Type> x (Const16 <x.Type> [d])))
+(OrB ((Less|Leq)8  (Const8  [c]) x) (Less8  x (Const8  [d]))) && c >= d -> ((Less|Leq)8U  (Const8  <x.Type> [c-d]) (Sub8  <x.Type> x (Const8  <x.Type> [d])))
+
+// signed integer range: ( c (<|<=) x || x <= d ) -> ( unsigned(c-(d+1)) (<|<=) unsigned(x-(d+1)) )
+(OrB ((Less|Leq)64 (Const64 [c]) x) (Leq64 x (Const64 [d]))) && c >= d+1 && int64(d+1) > int64(d) -> ((Less|Leq)64U (Const64 <x.Type> [c-d-1]) (Sub64 <x.Type> x (Const64 <x.Type> [d+1])))
+(OrB ((Less|Leq)32 (Const32 [c]) x) (Leq32 x (Const32 [d]))) && c >= d+1 && int32(d+1) > int32(d) -> ((Less|Leq)32U (Const32 <x.Type> [c-d-1]) (Sub32 <x.Type> x (Const32 <x.Type> [d+1])))
+(OrB ((Less|Leq)16 (Const16 [c]) x) (Leq16 x (Const16 [d]))) && c >= d+1 && int16(d+1) > int16(d) -> ((Less|Leq)16U (Const16 <x.Type> [c-d-1]) (Sub16 <x.Type> x (Const16 <x.Type> [d+1])))
+(OrB ((Less|Leq)8  (Const8  [c]) x) (Leq8  x (Const8  [d]))) && c >= d+1 && int8(d+1)  > int8(d)  -> ((Less|Leq)8U  (Const8  <x.Type> [c-d-1]) (Sub8  <x.Type> x (Const8  <x.Type> [d+1])))
+
+// unsigned integer range: ( c (<|<=) x || x < d ) -> ( c-d (<|<=) x-d )
+(OrB ((Less|Leq)64U (Const64 [c]) x) (Less64U x (Const64 [d]))) && uint64(c) >= uint64(d) -> ((Less|Leq)64U (Const64 <x.Type>               [c-d]) (Sub64 <x.Type> x (Const64 <x.Type> [d])))
+(OrB ((Less|Leq)32U (Const32 [c]) x) (Less32U x (Const32 [d]))) && uint32(c) >= uint32(d) -> ((Less|Leq)32U (Const32 <x.Type> [int64(int32(c-d))]) (Sub32 <x.Type> x (Const32 <x.Type> [d])))
+(OrB ((Less|Leq)16U (Const16 [c]) x) (Less16U x (Const16 [d]))) && uint16(c) >= uint16(d) -> ((Less|Leq)16U (Const16 <x.Type> [int64(int16(c-d))]) (Sub16 <x.Type> x (Const16 <x.Type> [d])))
+(OrB ((Less|Leq)8U  (Const8  [c]) x) (Less8U  x (Const8  [d]))) && uint8(c)  >= uint8(d)  -> ((Less|Leq)8U  (Const8  <x.Type> [int64( int8(c-d))]) (Sub8  <x.Type> x (Const8  <x.Type> [d])))
+
+// unsigned integer range: ( c (<|<=) x || x <= d ) -> ( c-(d+1) (<|<=) x-(d+1) )
+(OrB ((Less|Leq)64U (Const64 [c]) x) (Leq64U x (Const64 [d]))) && uint64(c) >= uint64(d+1) && uint64(d+1) > uint64(d) -> ((Less|Leq)64U (Const64 <x.Type>               [c-d-1]) (Sub64 <x.Type> x (Const64 <x.Type> [d+1])))
+(OrB ((Less|Leq)32U (Const32 [c]) x) (Leq32U x (Const32 [d]))) && uint32(c) >= uint32(d+1) && uint32(d+1) > uint32(d) -> ((Less|Leq)32U (Const32 <x.Type> [int64(int32(c-d-1))]) (Sub32 <x.Type> x (Const32 <x.Type> [int64(int32(d+1))])))
+(OrB ((Less|Leq)16U (Const16 [c]) x) (Leq16U x (Const16 [d]))) && uint16(c) >= uint16(d+1) && uint16(d+1) > uint16(d) -> ((Less|Leq)16U (Const16 <x.Type> [int64(int16(c-d-1))]) (Sub16 <x.Type> x (Const16 <x.Type> [int64(int16(d+1))])))
+(OrB ((Less|Leq)8U  (Const8  [c]) x) (Leq8U  x (Const8  [d]))) && uint8(c)  >= uint8(d+1)  && uint8(d+1)  > uint8(d)  -> ((Less|Leq)8U  (Const8  <x.Type> [int64( int8(c-d-1))]) (Sub8  <x.Type> x (Const8  <x.Type> [int64( int8(d+1))])))
+
 // Canonicalize x-const to x+(-const)
 (Sub64 x (Const64 <t> [c])) && x.Op != OpConst64 -> (Add64 (Const64 <t> [-c]) x)
 (Sub32 x (Const32 <t> [c])) && x.Op != OpConst32 -> (Add32 (Const32 <t> [int64(int32(-c))]) x)

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

@@ -87,7 +87,7 @@ func TestNilcheckSimple(t *testing.T) {
 	nilcheckelim(fun.f)
 
 	// clean up the removed nil check
-	fusePlain(fun.f)
+	fuse(fun.f, fuseTypePlain)
 	deadcode(fun.f)
 
 	CheckFunc(fun.f)
@@ -124,7 +124,7 @@ func TestNilcheckDomOrder(t *testing.T) {
 	nilcheckelim(fun.f)
 
 	// clean up the removed nil check
-	fusePlain(fun.f)
+	fuse(fun.f, fuseTypePlain)
 	deadcode(fun.f)
 
 	CheckFunc(fun.f)
@@ -157,7 +157,7 @@ func TestNilcheckAddr(t *testing.T) {
 	nilcheckelim(fun.f)
 
 	// clean up the removed nil check
-	fusePlain(fun.f)
+	fuse(fun.f, fuseTypePlain)
 	deadcode(fun.f)
 
 	CheckFunc(fun.f)
@@ -191,7 +191,7 @@ func TestNilcheckAddPtr(t *testing.T) {
 	nilcheckelim(fun.f)
 
 	// clean up the removed nil check
-	fusePlain(fun.f)
+	fuse(fun.f, fuseTypePlain)
 	deadcode(fun.f)
 
 	CheckFunc(fun.f)
@@ -235,7 +235,7 @@ func TestNilcheckPhi(t *testing.T) {
 	nilcheckelim(fun.f)
 
 	// clean up the removed nil check
-	fusePlain(fun.f)
+	fuse(fun.f, fuseTypePlain)
 	deadcode(fun.f)
 
 	CheckFunc(fun.f)
@@ -276,7 +276,7 @@ func TestNilcheckKeepRemove(t *testing.T) {
 	nilcheckelim(fun.f)
 
 	// clean up the removed nil check
-	fusePlain(fun.f)
+	fuse(fun.f, fuseTypePlain)
 	deadcode(fun.f)
 
 	CheckFunc(fun.f)
@@ -323,7 +323,7 @@ func TestNilcheckInFalseBranch(t *testing.T) {
 	nilcheckelim(fun.f)
 
 	// clean up the removed nil check
-	fusePlain(fun.f)
+	fuse(fun.f, fuseTypePlain)
 	deadcode(fun.f)
 
 	CheckFunc(fun.f)
@@ -374,7 +374,7 @@ func TestNilcheckUser(t *testing.T) {
 	nilcheckelim(fun.f)
 
 	// clean up the removed nil check
-	fusePlain(fun.f)
+	fuse(fun.f, fuseTypePlain)
 	deadcode(fun.f)
 
 	CheckFunc(fun.f)
@@ -418,7 +418,7 @@ func TestNilcheckBug(t *testing.T) {
 	nilcheckelim(fun.f)
 
 	// clean up the removed nil check
-	fusePlain(fun.f)
+	fuse(fun.f, fuseTypePlain)
 	deadcode(fun.f)
 
 	CheckFunc(fun.f)

test/codegen/fuse.go

@@ -0,0 +1,197 @@
+// asmcheck
+
+// Copyright 2019 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
+
+// Notes:
+// - these examples use channels to provide a source of
+//   unknown values that cannot be optimized away
+// - these examples use for loops to force branches
+//   backward (predicted taken)
+
+// ---------------------------------- //
+// signed integer range (conjunction) //
+// ---------------------------------- //
+
+func si1c(c <-chan int64) {
+	// amd64:"CMPQ\t.+, [$]256"
+	// s390x:"CLGIJ\t[$]12, R[0-9]+, [$]255"
+	for x := <-c; x >= 0 && x < 256; x = <-c {
+	}
+}
+
+func si2c(c <-chan int32) {
+	// amd64:"CMPL\t.+, [$]256"
+	// s390x:"CLIJ\t[$]12, R[0-9]+, [$]255"
+	for x := <-c; x >= 0 && x < 256; x = <-c {
+	}
+}
+
+func si3c(c <-chan int16) {
+	// amd64:"CMPW\t.+, [$]256"
+	// s390x:"CLIJ\t[$]12, R[0-9]+, [$]255"
+	for x := <-c; x >= 0 && x < 256; x = <-c {
+	}
+}
+
+func si4c(c <-chan int8) {
+	// amd64:"CMPB\t.+, [$]10"
+	// s390x:"CLIJ\t[$]4, R[0-9]+, [$]10"
+	for x := <-c; x >= 0 && x < 10; x = <-c {
+	}
+}
+
+func si5c(c <-chan int64) {
+	// amd64:"CMPQ\t.+, [$]251","ADDQ\t[$]-5,"
+	// s390x:"CLGIJ\t[$]4, R[0-9]+, [$]251","ADD\t[$]-5,"
+	for x := <-c; x < 256 && x > 4; x = <-c {
+	}
+}
+
+func si6c(c <-chan int32) {
+	// amd64:"CMPL\t.+, [$]255","DECL\t"
+	// s390x:"CLIJ\t[$]12, R[0-9]+, [$]255","ADDW\t[$]-1,"
+	for x := <-c; x > 0 && x <= 256; x = <-c {
+	}
+}
+
+func si7c(c <-chan int16) {
+	// amd64:"CMPW\t.+, [$]60","ADDL\t[$]10,"
+	// s390x:"CLIJ\t[$]12, R[0-9]+, [$]60","ADDW\t[$]10,"
+	for x := <-c; x >= -10 && x <= 50; x = <-c {
+	}
+}
+
+func si8c(c <-chan int8) {
+	// amd64:"CMPB\t.+, [$]126","ADDL\t[$]126,"
+	// s390x:"CLIJ\t[$]4, R[0-9]+, [$]126","ADDW\t[$]126,"
+	for x := <-c; x >= -126 && x < 0; x = <-c {
+	}
+}
+
+// ---------------------------------- //
+// signed integer range (disjunction) //
+// ---------------------------------- //
+
+func si1d(c <-chan int64) {
+	// amd64:"CMPQ\t.+, [$]256"
+	// s390x:"CLGIJ\t[$]2, R[0-9]+, [$]255"
+	for x := <-c; x < 0 || x >= 256; x = <-c {
+	}
+}
+
+func si2d(c <-chan int32) {
+	// amd64:"CMPL\t.+, [$]256"
+	// s390x:"CLIJ\t[$]2, R[0-9]+, [$]255"
+	for x := <-c; x < 0 || x >= 256; x = <-c {
+	}
+}
+
+func si3d(c <-chan int16) {
+	// amd64:"CMPW\t.+, [$]256"
+	// s390x:"CLIJ\t[$]2, R[0-9]+, [$]255"
+	for x := <-c; x < 0 || x >= 256; x = <-c {
+	}
+}
+
+func si4d(c <-chan int8) {
+	// amd64:"CMPB\t.+, [$]10"
+	// s390x:"CLIJ\t[$]10, R[0-9]+, [$]10"
+	for x := <-c; x < 0 || x >= 10; x = <-c {
+	}
+}
+
+func si5d(c <-chan int64) {
+	// amd64:"CMPQ\t.+, [$]251","ADDQ\t[$]-5,"
+	// s390x:"CLGIJ\t[$]10, R[0-9]+, [$]251","ADD\t[$]-5,"
+	for x := <-c; x >= 256 || x <= 4; x = <-c {
+	}
+}
+
+func si6d(c <-chan int32) {
+	// amd64:"CMPL\t.+, [$]255","DECL\t"
+	// s390x:"CLIJ\t[$]2, R[0-9]+, [$]255","ADDW\t[$]-1,"
+	for x := <-c; x <= 0 || x > 256; x = <-c {
+	}
+}
+
+func si7d(c <-chan int16) {
+	// amd64:"CMPW\t.+, [$]60","ADDL\t[$]10,"
+	// s390x:"CLIJ\t[$]2, R[0-9]+, [$]60","ADDW\t[$]10,"
+	for x := <-c; x < -10 || x > 50; x = <-c {
+	}
+}
+
+func si8d(c <-chan int8) {
+	// amd64:"CMPB\t.+, [$]126","ADDL\t[$]126,"
+	// s390x:"CLIJ\t[$]10, R[0-9]+, [$]126","ADDW\t[$]126,"
+	for x := <-c; x < -126 || x >= 0; x = <-c {
+	}
+}
+
+// ------------------------------------ //
+// unsigned integer range (conjunction) //
+// ------------------------------------ //
+
+func ui1c(c <-chan uint64) {
+	// amd64:"CMPQ\t.+, [$]251","ADDQ\t[$]-5,"
+	// s390x:"CLGIJ\t[$]4, R[0-9]+, [$]251","ADD\t[$]-5,"
+	for x := <-c; x < 256 && x > 4; x = <-c {
+	}
+}
+
+func ui2c(c <-chan uint32) {
+	// amd64:"CMPL\t.+, [$]255","DECL\t"
+	// s390x:"CLIJ\t[$]12, R[0-9]+, [$]255","ADDW\t[$]-1,"
+	for x := <-c; x > 0 && x <= 256; x = <-c {
+	}
+}
+
+func ui3c(c <-chan uint16) {
+	// amd64:"CMPW\t.+, [$]40","ADDL\t[$]-10,"
+	// s390x:"CLIJ\t[$]12, R[0-9]+, [$]40","ADDW\t[$]-10,"
+	for x := <-c; x >= 10 && x <= 50; x = <-c {
+	}
+}
+
+func ui4c(c <-chan uint8) {
+	// amd64:"CMPB\t.+, [$]2","ADDL\t[$]-126,"
+	// s390x:"CLIJ\t[$]4, R[0-9]+, [$]2","ADDW\t[$]-126,"
+	for x := <-c; x >= 126 && x < 128; x = <-c {
+	}
+}
+
+// ------------------------------------ //
+// unsigned integer range (disjunction) //
+// ------------------------------------ //
+
+func ui1d(c <-chan uint64) {
+	// amd64:"CMPQ\t.+, [$]251","ADDQ\t[$]-5,"
+	// s390x:"CLGIJ\t[$]10, R[0-9]+, [$]251","ADD\t[$]-5,"
+	for x := <-c; x >= 256 || x <= 4; x = <-c {
+	}
+}
+
+func ui2d(c <-chan uint32) {
+	// amd64:"CMPL\t.+, [$]254","ADDL\t[$]-2,"
+	// s390x:"CLIJ\t[$]2, R[0-9]+, [$]254","ADDW\t[$]-2,"
+	for x := <-c; x <= 1 || x > 256; x = <-c {
+	}
+}
+
+func ui3d(c <-chan uint16) {
+	// amd64:"CMPW\t.+, [$]40","ADDL\t[$]-10,"
+	// s390x:"CLIJ\t[$]2, R[0-9]+, [$]40","ADDW\t[$]-10,"
+	for x := <-c; x < 10 || x > 50; x = <-c {
+	}
+}
+
+func ui4d(c <-chan uint8) {
+	// amd64:"CMPB\t.+, [$]2","ADDL\t[$]-126,"
+	// s390x:"CLIJ\t[$]10, R[0-9]+, [$]2","ADDW\t[$]-126,"
+	for x := <-c; x < 126 || x >= 128; x = <-c {
+	}
+}