qbit/go
1
0
Fork 0
mirror of https://github.com/golang/go synced 2024-11-15 01:20:28 -07:00
Code Releases Activity
aec9b916a2
go/test/inline_big.go

1030 lines
13 KiB
Go
Raw Normal View History

cmd/compile: set stricter inlining threshold in large functions If we're compiling a large function, be more picky about how big the function we're inlining is. If the function is >5000 nodes, we lower the inlining threshold from a cost of 80 to 20. Turns out reflect.Value's cost is exactly 80. That's the function at issue in #26546. 20 was chosen as a proxy for "inlined body is smaller than the call would be". Simple functions still get inlined, like this one at cost 7: func ifaceIndir(t *rtype) bool { return t.kind&kindDirectIface == 0 } 5000 nodes was chosen as the big function size. Here are all the 5000+ node (~~1000+ lines) functions in the stdlib: 5187 cmd/internal/obj/arm (*ctxt5).asmout 6879 cmd/internal/obj/s390x (*ctxtz).asmout 6567 cmd/internal/obj/ppc64 (*ctxt9).asmout 9643 cmd/internal/obj/arm64 (*ctxt7).asmout 5042 cmd/internal/obj/x86 (*AsmBuf).doasm 8768 cmd/compile/internal/ssa rewriteBlockAMD64 8878 cmd/compile/internal/ssa rewriteBlockARM 8344 cmd/compile/internal/ssa rewriteValueARM64_OpARM64OR_20 7916 cmd/compile/internal/ssa rewriteValueARM64_OpARM64OR_30 5427 cmd/compile/internal/ssa rewriteBlockARM64 5126 cmd/compile/internal/ssa rewriteValuePPC64_OpPPC64OR_50 6152 cmd/compile/internal/ssa rewriteValuePPC64_OpPPC64OR_60 6412 cmd/compile/internal/ssa rewriteValuePPC64_OpPPC64OR_70 6486 cmd/compile/internal/ssa rewriteValuePPC64_OpPPC64OR_80 6534 cmd/compile/internal/ssa rewriteValuePPC64_OpPPC64OR_90 6534 cmd/compile/internal/ssa rewriteValuePPC64_OpPPC64OR_100 6534 cmd/compile/internal/ssa rewriteValuePPC64_OpPPC64OR_110 6675 cmd/compile/internal/gc typecheck1 5433 cmd/compile/internal/gc walkexpr 14070 cmd/vendor/golang.org/x/arch/arm64/arm64asm decodeArg There are a lot more smaller (~1000 node) functions in the stdlib. The function in #26546 has 12477 nodes. At some point it might be nice to have a better heuristic for "inlined body is smaller than the call", a non-cliff way to scale down the cost as the function gets bigger, doing cheaper inlined calls first, etc. All that can wait for another release. I'd like to do this CL for 1.11. Fixes #26546 Update #17566 Change-Id: Idda13020e46ec2b28d79a17217f44b189f8139ac Reviewed-on: https://go-review.googlesource.com/125516 Run-TryBot: Keith Randall <khr@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: David Chase <drchase@google.com>
2018-07-23 14:09:48 -06:00
// errorcheck -0 -m=2
// 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.
// Test that we restrict inlining into very large functions.
// See issue #26546.
package foo
cmd/compile: restore zero-copy string->[]byte optimization This CL implements the remainder of the zero-copy string->[]byte conversion optimization initially attempted in go.dev/cl/520395, but fixes the tracking of mutations due to ODEREF/ODOTPTR assignments, and adds more comprehensive tests that I should have included originally. However, this CL also keeps it behind the -d=zerocopy flag. The next CL will enable it by default (for easier rollback). Updates #2205. Change-Id: Ic330260099ead27fc00e2680a59c6ff23cb63c2b Reviewed-on: https://go-review.googlesource.com/c/go/+/520599 Auto-Submit: Matthew Dempsky <mdempsky@google.com> TryBot-Result: Gopher Robot <gobot@golang.org> Reviewed-by: Than McIntosh <thanm@google.com> Run-TryBot: Matthew Dempsky <mdempsky@google.com> Reviewed-by: Cuong Manh Le <cuong.manhle.vn@gmail.com>
2023-08-17 15:15:04 -06:00
func small(a []int) int { // ERROR "can inline small with cost .* as:.*" "a does not escape"
cmd/compile: set stricter inlining threshold in large functions If we're compiling a large function, be more picky about how big the function we're inlining is. If the function is >5000 nodes, we lower the inlining threshold from a cost of 80 to 20. Turns out reflect.Value's cost is exactly 80. That's the function at issue in #26546. 20 was chosen as a proxy for "inlined body is smaller than the call would be". Simple functions still get inlined, like this one at cost 7: func ifaceIndir(t *rtype) bool { return t.kind&kindDirectIface == 0 } 5000 nodes was chosen as the big function size. Here are all the 5000+ node (~~1000+ lines) functions in the stdlib: 5187 cmd/internal/obj/arm (*ctxt5).asmout 6879 cmd/internal/obj/s390x (*ctxtz).asmout 6567 cmd/internal/obj/ppc64 (*ctxt9).asmout 9643 cmd/internal/obj/arm64 (*ctxt7).asmout 5042 cmd/internal/obj/x86 (*AsmBuf).doasm 8768 cmd/compile/internal/ssa rewriteBlockAMD64 8878 cmd/compile/internal/ssa rewriteBlockARM 8344 cmd/compile/internal/ssa rewriteValueARM64_OpARM64OR_20 7916 cmd/compile/internal/ssa rewriteValueARM64_OpARM64OR_30 5427 cmd/compile/internal/ssa rewriteBlockARM64 5126 cmd/compile/internal/ssa rewriteValuePPC64_OpPPC64OR_50 6152 cmd/compile/internal/ssa rewriteValuePPC64_OpPPC64OR_60 6412 cmd/compile/internal/ssa rewriteValuePPC64_OpPPC64OR_70 6486 cmd/compile/internal/ssa rewriteValuePPC64_OpPPC64OR_80 6534 cmd/compile/internal/ssa rewriteValuePPC64_OpPPC64OR_90 6534 cmd/compile/internal/ssa rewriteValuePPC64_OpPPC64OR_100 6534 cmd/compile/internal/ssa rewriteValuePPC64_OpPPC64OR_110 6675 cmd/compile/internal/gc typecheck1 5433 cmd/compile/internal/gc walkexpr 14070 cmd/vendor/golang.org/x/arch/arm64/arm64asm decodeArg There are a lot more smaller (~1000 node) functions in the stdlib. The function in #26546 has 12477 nodes. At some point it might be nice to have a better heuristic for "inlined body is smaller than the call", a non-cliff way to scale down the cost as the function gets bigger, doing cheaper inlined calls first, etc. All that can wait for another release. I'd like to do this CL for 1.11. Fixes #26546 Update #17566 Change-Id: Idda13020e46ec2b28d79a17217f44b189f8139ac Reviewed-on: https://go-review.googlesource.com/125516 Run-TryBot: Keith Randall <khr@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: David Chase <drchase@google.com>
2018-07-23 14:09:48 -06:00
// Cost 16 body (need cost < 20).
// See cmd/compile/internal/gc/inl.go:inlineBigFunction*
return a[0] + a[1] + a[2] + a[3]
}
cmd/compile: restore zero-copy string->[]byte optimization This CL implements the remainder of the zero-copy string->[]byte conversion optimization initially attempted in go.dev/cl/520395, but fixes the tracking of mutations due to ODEREF/ODOTPTR assignments, and adds more comprehensive tests that I should have included originally. However, this CL also keeps it behind the -d=zerocopy flag. The next CL will enable it by default (for easier rollback). Updates #2205. Change-Id: Ic330260099ead27fc00e2680a59c6ff23cb63c2b Reviewed-on: https://go-review.googlesource.com/c/go/+/520599 Auto-Submit: Matthew Dempsky <mdempsky@google.com> TryBot-Result: Gopher Robot <gobot@golang.org> Reviewed-by: Than McIntosh <thanm@google.com> Run-TryBot: Matthew Dempsky <mdempsky@google.com> Reviewed-by: Cuong Manh Le <cuong.manhle.vn@gmail.com>
2023-08-17 15:15:04 -06:00
func medium(a []int) int { // ERROR "can inline medium with cost .* as:.*" "a does not escape"
cmd/compile: set stricter inlining threshold in large functions If we're compiling a large function, be more picky about how big the function we're inlining is. If the function is >5000 nodes, we lower the inlining threshold from a cost of 80 to 20. Turns out reflect.Value's cost is exactly 80. That's the function at issue in #26546. 20 was chosen as a proxy for "inlined body is smaller than the call would be". Simple functions still get inlined, like this one at cost 7: func ifaceIndir(t *rtype) bool { return t.kind&kindDirectIface == 0 } 5000 nodes was chosen as the big function size. Here are all the 5000+ node (~~1000+ lines) functions in the stdlib: 5187 cmd/internal/obj/arm (*ctxt5).asmout 6879 cmd/internal/obj/s390x (*ctxtz).asmout 6567 cmd/internal/obj/ppc64 (*ctxt9).asmout 9643 cmd/internal/obj/arm64 (*ctxt7).asmout 5042 cmd/internal/obj/x86 (*AsmBuf).doasm 8768 cmd/compile/internal/ssa rewriteBlockAMD64 8878 cmd/compile/internal/ssa rewriteBlockARM 8344 cmd/compile/internal/ssa rewriteValueARM64_OpARM64OR_20 7916 cmd/compile/internal/ssa rewriteValueARM64_OpARM64OR_30 5427 cmd/compile/internal/ssa rewriteBlockARM64 5126 cmd/compile/internal/ssa rewriteValuePPC64_OpPPC64OR_50 6152 cmd/compile/internal/ssa rewriteValuePPC64_OpPPC64OR_60 6412 cmd/compile/internal/ssa rewriteValuePPC64_OpPPC64OR_70 6486 cmd/compile/internal/ssa rewriteValuePPC64_OpPPC64OR_80 6534 cmd/compile/internal/ssa rewriteValuePPC64_OpPPC64OR_90 6534 cmd/compile/internal/ssa rewriteValuePPC64_OpPPC64OR_100 6534 cmd/compile/internal/ssa rewriteValuePPC64_OpPPC64OR_110 6675 cmd/compile/internal/gc typecheck1 5433 cmd/compile/internal/gc walkexpr 14070 cmd/vendor/golang.org/x/arch/arm64/arm64asm decodeArg There are a lot more smaller (~1000 node) functions in the stdlib. The function in #26546 has 12477 nodes. At some point it might be nice to have a better heuristic for "inlined body is smaller than the call", a non-cliff way to scale down the cost as the function gets bigger, doing cheaper inlined calls first, etc. All that can wait for another release. I'd like to do this CL for 1.11. Fixes #26546 Update #17566 Change-Id: Idda13020e46ec2b28d79a17217f44b189f8139ac Reviewed-on: https://go-review.googlesource.com/125516 Run-TryBot: Keith Randall <khr@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: David Chase <drchase@google.com>
2018-07-23 14:09:48 -06:00
// Cost 32 body (need cost > 20 and cost < 80).
// See cmd/compile/internal/gc/inl.go:inlineBigFunction*
return a[0] + a[1] + a[2] + a[3] + a[4] + a[5] + a[6] + a[7]
}
cmd/compile: restore zero-copy string->[]byte optimization This CL implements the remainder of the zero-copy string->[]byte conversion optimization initially attempted in go.dev/cl/520395, but fixes the tracking of mutations due to ODEREF/ODOTPTR assignments, and adds more comprehensive tests that I should have included originally. However, this CL also keeps it behind the -d=zerocopy flag. The next CL will enable it by default (for easier rollback). Updates #2205. Change-Id: Ic330260099ead27fc00e2680a59c6ff23cb63c2b Reviewed-on: https://go-review.googlesource.com/c/go/+/520599 Auto-Submit: Matthew Dempsky <mdempsky@google.com> TryBot-Result: Gopher Robot <gobot@golang.org> Reviewed-by: Than McIntosh <thanm@google.com> Run-TryBot: Matthew Dempsky <mdempsky@google.com> Reviewed-by: Cuong Manh Le <cuong.manhle.vn@gmail.com>
2023-08-17 15:15:04 -06:00
func f(a []int) int { // ERROR "cannot inline f:.*" "a does not escape" "function f considered 'big'"
cmd/compile: set stricter inlining threshold in large functions If we're compiling a large function, be more picky about how big the function we're inlining is. If the function is >5000 nodes, we lower the inlining threshold from a cost of 80 to 20. Turns out reflect.Value's cost is exactly 80. That's the function at issue in #26546. 20 was chosen as a proxy for "inlined body is smaller than the call would be". Simple functions still get inlined, like this one at cost 7: func ifaceIndir(t *rtype) bool { return t.kind&kindDirectIface == 0 } 5000 nodes was chosen as the big function size. Here are all the 5000+ node (~~1000+ lines) functions in the stdlib: 5187 cmd/internal/obj/arm (*ctxt5).asmout 6879 cmd/internal/obj/s390x (*ctxtz).asmout 6567 cmd/internal/obj/ppc64 (*ctxt9).asmout 9643 cmd/internal/obj/arm64 (*ctxt7).asmout 5042 cmd/internal/obj/x86 (*AsmBuf).doasm 8768 cmd/compile/internal/ssa rewriteBlockAMD64 8878 cmd/compile/internal/ssa rewriteBlockARM 8344 cmd/compile/internal/ssa rewriteValueARM64_OpARM64OR_20 7916 cmd/compile/internal/ssa rewriteValueARM64_OpARM64OR_30 5427 cmd/compile/internal/ssa rewriteBlockARM64 5126 cmd/compile/internal/ssa rewriteValuePPC64_OpPPC64OR_50 6152 cmd/compile/internal/ssa rewriteValuePPC64_OpPPC64OR_60 6412 cmd/compile/internal/ssa rewriteValuePPC64_OpPPC64OR_70 6486 cmd/compile/internal/ssa rewriteValuePPC64_OpPPC64OR_80 6534 cmd/compile/internal/ssa rewriteValuePPC64_OpPPC64OR_90 6534 cmd/compile/internal/ssa rewriteValuePPC64_OpPPC64OR_100 6534 cmd/compile/internal/ssa rewriteValuePPC64_OpPPC64OR_110 6675 cmd/compile/internal/gc typecheck1 5433 cmd/compile/internal/gc walkexpr 14070 cmd/vendor/golang.org/x/arch/arm64/arm64asm decodeArg There are a lot more smaller (~1000 node) functions in the stdlib. The function in #26546 has 12477 nodes. At some point it might be nice to have a better heuristic for "inlined body is smaller than the call", a non-cliff way to scale down the cost as the function gets bigger, doing cheaper inlined calls first, etc. All that can wait for another release. I'd like to do this CL for 1.11. Fixes #26546 Update #17566 Change-Id: Idda13020e46ec2b28d79a17217f44b189f8139ac Reviewed-on: https://go-review.googlesource.com/125516 Run-TryBot: Keith Randall <khr@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: David Chase <drchase@google.com>
2018-07-23 14:09:48 -06:00
// Add lots of nodes to f's body. We need >5000.
// See cmd/compile/internal/gc/inl.go:inlineBigFunction*
a[0] = 0
a[1] = 0
a[2] = 0
a[3] = 0
a[4] = 0
a[5] = 0
a[6] = 0
a[7] = 0
a[8] = 0
a[9] = 0
a[10] = 0
a[11] = 0
a[12] = 0
a[13] = 0
a[14] = 0
a[15] = 0
a[16] = 0
a[17] = 0
a[18] = 0
a[19] = 0
a[20] = 0
a[21] = 0
a[22] = 0
a[23] = 0
a[24] = 0
a[25] = 0
a[26] = 0
a[27] = 0
a[28] = 0
a[29] = 0
a[30] = 0
a[31] = 0
a[32] = 0
a[33] = 0
a[34] = 0
a[35] = 0
a[36] = 0
a[37] = 0
a[38] = 0
a[39] = 0
a[40] = 0
a[41] = 0
a[42] = 0
a[43] = 0
a[44] = 0
a[45] = 0
a[46] = 0
a[47] = 0
a[48] = 0
a[49] = 0
a[50] = 0
a[51] = 0
a[52] = 0
a[53] = 0
a[54] = 0
a[55] = 0
a[56] = 0
a[57] = 0
a[58] = 0
a[59] = 0
a[60] = 0
a[61] = 0
a[62] = 0
a[63] = 0
a[64] = 0
a[65] = 0
a[66] = 0
a[67] = 0
a[68] = 0
a[69] = 0
a[70] = 0
a[71] = 0
a[72] = 0
a[73] = 0
a[74] = 0
a[75] = 0
a[76] = 0
a[77] = 0
a[78] = 0
a[79] = 0
a[80] = 0
a[81] = 0
a[82] = 0
a[83] = 0
a[84] = 0
a[85] = 0
a[86] = 0
a[87] = 0
a[88] = 0
a[89] = 0
a[90] = 0
a[91] = 0
a[92] = 0
a[93] = 0
a[94] = 0
a[95] = 0
a[96] = 0
a[97] = 0
a[98] = 0
a[99] = 0
a[100] = 0
a[101] = 0
a[102] = 0
a[103] = 0
a[104] = 0
a[105] = 0
a[106] = 0
a[107] = 0
a[108] = 0
a[109] = 0
a[110] = 0
a[111] = 0
a[112] = 0
a[113] = 0
a[114] = 0
a[115] = 0
a[116] = 0
a[117] = 0
a[118] = 0
a[119] = 0
a[120] = 0
a[121] = 0
a[122] = 0
a[123] = 0
a[124] = 0
a[125] = 0
a[126] = 0
a[127] = 0
a[128] = 0
a[129] = 0
a[130] = 0
a[131] = 0
a[132] = 0
a[133] = 0
a[134] = 0
a[135] = 0
a[136] = 0
a[137] = 0
a[138] = 0
a[139] = 0
a[140] = 0
a[141] = 0
a[142] = 0
a[143] = 0
a[144] = 0
a[145] = 0
a[146] = 0
a[147] = 0
a[148] = 0
a[149] = 0
a[150] = 0
a[151] = 0
a[152] = 0
a[153] = 0
a[154] = 0
a[155] = 0
a[156] = 0
a[157] = 0
a[158] = 0
a[159] = 0
a[160] = 0
a[161] = 0
a[162] = 0
a[163] = 0
a[164] = 0
a[165] = 0
a[166] = 0
a[167] = 0
a[168] = 0
a[169] = 0
a[170] = 0
a[171] = 0
a[172] = 0
a[173] = 0
a[174] = 0
a[175] = 0
a[176] = 0
a[177] = 0
a[178] = 0
a[179] = 0
a[180] = 0
a[181] = 0
a[182] = 0
a[183] = 0
a[184] = 0
a[185] = 0
a[186] = 0
a[187] = 0
a[188] = 0
a[189] = 0
a[190] = 0
a[191] = 0
a[192] = 0
a[193] = 0
a[194] = 0
a[195] = 0
a[196] = 0
a[197] = 0
a[198] = 0
a[199] = 0
a[200] = 0
a[201] = 0
a[202] = 0
a[203] = 0
a[204] = 0
a[205] = 0
a[206] = 0
a[207] = 0
a[208] = 0
a[209] = 0
a[210] = 0
a[211] = 0
a[212] = 0
a[213] = 0
a[214] = 0
a[215] = 0
a[216] = 0
a[217] = 0
a[218] = 0
a[219] = 0
a[220] = 0
a[221] = 0
a[222] = 0
a[223] = 0
a[224] = 0
a[225] = 0
a[226] = 0
a[227] = 0
a[228] = 0
a[229] = 0
a[230] = 0
a[231] = 0
a[232] = 0
a[233] = 0
a[234] = 0
a[235] = 0
a[236] = 0
a[237] = 0
a[238] = 0
a[239] = 0
a[240] = 0
a[241] = 0
a[242] = 0
a[243] = 0
a[244] = 0
a[245] = 0
a[246] = 0
a[247] = 0
a[248] = 0
a[249] = 0
a[250] = 0
a[251] = 0
a[252] = 0
a[253] = 0
a[254] = 0
a[255] = 0
a[256] = 0
a[257] = 0
a[258] = 0
a[259] = 0
a[260] = 0
a[261] = 0
a[262] = 0
a[263] = 0
a[264] = 0
a[265] = 0
a[266] = 0
a[267] = 0
a[268] = 0
a[269] = 0
a[270] = 0
a[271] = 0
a[272] = 0
a[273] = 0
a[274] = 0
a[275] = 0
a[276] = 0
a[277] = 0
a[278] = 0
a[279] = 0
a[280] = 0
a[281] = 0
a[282] = 0
a[283] = 0
a[284] = 0
a[285] = 0
a[286] = 0
a[287] = 0
a[288] = 0
a[289] = 0
a[290] = 0
a[291] = 0
a[292] = 0
a[293] = 0
a[294] = 0
a[295] = 0
a[296] = 0
a[297] = 0
a[298] = 0
a[299] = 0
a[300] = 0
a[301] = 0
a[302] = 0
a[303] = 0
a[304] = 0
a[305] = 0
a[306] = 0
a[307] = 0
a[308] = 0
a[309] = 0
a[310] = 0
a[311] = 0
a[312] = 0
a[313] = 0
a[314] = 0
a[315] = 0
a[316] = 0
a[317] = 0
a[318] = 0
a[319] = 0
a[320] = 0
a[321] = 0
a[322] = 0
a[323] = 0
a[324] = 0
a[325] = 0
a[326] = 0
a[327] = 0
a[328] = 0
a[329] = 0
a[330] = 0
a[331] = 0
a[332] = 0
a[333] = 0
a[334] = 0
a[335] = 0
a[336] = 0
a[337] = 0
a[338] = 0
a[339] = 0
a[340] = 0
a[341] = 0
a[342] = 0
a[343] = 0
a[344] = 0
a[345] = 0
a[346] = 0
a[347] = 0
a[348] = 0
a[349] = 0
a[350] = 0
a[351] = 0
a[352] = 0
a[353] = 0
a[354] = 0
a[355] = 0
a[356] = 0
a[357] = 0
a[358] = 0
a[359] = 0
a[360] = 0
a[361] = 0
a[362] = 0
a[363] = 0
a[364] = 0
a[365] = 0
a[366] = 0
a[367] = 0
a[368] = 0
a[369] = 0
a[370] = 0
a[371] = 0
a[372] = 0
a[373] = 0
a[374] = 0
a[375] = 0
a[376] = 0
a[377] = 0
a[378] = 0
a[379] = 0
a[380] = 0
a[381] = 0
a[382] = 0
a[383] = 0
a[384] = 0
a[385] = 0
a[386] = 0
a[387] = 0
a[388] = 0
a[389] = 0
a[390] = 0
a[391] = 0
a[392] = 0
a[393] = 0
a[394] = 0
a[395] = 0
a[396] = 0
a[397] = 0
a[398] = 0
a[399] = 0
a[400] = 0
a[401] = 0
a[402] = 0
a[403] = 0
a[404] = 0
a[405] = 0
a[406] = 0
a[407] = 0
a[408] = 0
a[409] = 0
a[410] = 0
a[411] = 0
a[412] = 0
a[413] = 0
a[414] = 0
a[415] = 0
a[416] = 0
a[417] = 0
a[418] = 0
a[419] = 0
a[420] = 0
a[421] = 0
a[422] = 0
a[423] = 0
a[424] = 0
a[425] = 0
a[426] = 0
a[427] = 0
a[428] = 0
a[429] = 0
a[430] = 0
a[431] = 0
a[432] = 0
a[433] = 0
a[434] = 0
a[435] = 0
a[436] = 0
a[437] = 0
a[438] = 0
a[439] = 0
a[440] = 0
a[441] = 0
a[442] = 0
a[443] = 0
a[444] = 0
a[445] = 0
a[446] = 0
a[447] = 0
a[448] = 0
a[449] = 0
a[450] = 0
a[451] = 0
a[452] = 0
a[453] = 0
a[454] = 0
a[455] = 0
a[456] = 0
a[457] = 0
a[458] = 0
a[459] = 0
a[460] = 0
a[461] = 0
a[462] = 0
a[463] = 0
a[464] = 0
a[465] = 0
a[466] = 0
a[467] = 0
a[468] = 0
a[469] = 0
a[470] = 0
a[471] = 0
a[472] = 0
a[473] = 0
a[474] = 0
a[475] = 0
a[476] = 0
a[477] = 0
a[478] = 0
a[479] = 0
a[480] = 0
a[481] = 0
a[482] = 0
a[483] = 0
a[484] = 0
a[485] = 0
a[486] = 0
a[487] = 0
a[488] = 0
a[489] = 0
a[490] = 0
a[491] = 0
a[492] = 0
a[493] = 0
a[494] = 0
a[495] = 0
a[496] = 0
a[497] = 0
a[498] = 0
a[499] = 0
a[500] = 0
a[501] = 0
a[502] = 0
a[503] = 0
a[504] = 0
a[505] = 0
a[506] = 0
a[507] = 0
a[508] = 0
a[509] = 0
a[510] = 0
a[511] = 0
a[512] = 0
a[513] = 0
a[514] = 0
a[515] = 0
a[516] = 0
a[517] = 0
a[518] = 0
a[519] = 0
a[520] = 0
a[521] = 0
a[522] = 0
a[523] = 0
a[524] = 0
a[525] = 0
a[526] = 0
a[527] = 0
a[528] = 0
a[529] = 0
a[530] = 0
a[531] = 0
a[532] = 0
a[533] = 0
a[534] = 0
a[535] = 0
a[536] = 0
a[537] = 0
a[538] = 0
a[539] = 0
a[540] = 0
a[541] = 0
a[542] = 0
a[543] = 0
a[544] = 0
a[545] = 0
a[546] = 0
a[547] = 0
a[548] = 0
a[549] = 0
a[550] = 0
a[551] = 0
a[552] = 0
a[553] = 0
a[554] = 0
a[555] = 0
a[556] = 0
a[557] = 0
a[558] = 0
a[559] = 0
a[560] = 0
a[561] = 0
a[562] = 0
a[563] = 0
a[564] = 0
a[565] = 0
a[566] = 0
a[567] = 0
a[568] = 0
a[569] = 0
a[570] = 0
a[571] = 0
a[572] = 0
a[573] = 0
a[574] = 0
a[575] = 0
a[576] = 0
a[577] = 0
a[578] = 0
a[579] = 0
a[580] = 0
a[581] = 0
a[582] = 0
a[583] = 0
a[584] = 0
a[585] = 0
a[586] = 0
a[587] = 0
a[588] = 0
a[589] = 0
a[590] = 0
a[591] = 0
a[592] = 0
a[593] = 0
a[594] = 0
a[595] = 0
a[596] = 0
a[597] = 0
a[598] = 0
a[599] = 0
a[600] = 0
a[601] = 0
a[602] = 0
a[603] = 0
a[604] = 0
a[605] = 0
a[606] = 0
a[607] = 0
a[608] = 0
a[609] = 0
a[610] = 0
a[611] = 0
a[612] = 0
a[613] = 0
a[614] = 0
a[615] = 0
a[616] = 0
a[617] = 0
a[618] = 0
a[619] = 0
a[620] = 0
a[621] = 0
a[622] = 0
a[623] = 0
a[624] = 0
a[625] = 0
a[626] = 0
a[627] = 0
a[628] = 0
a[629] = 0
a[630] = 0
a[631] = 0
a[632] = 0
a[633] = 0
a[634] = 0
a[635] = 0
a[636] = 0
a[637] = 0
a[638] = 0
a[639] = 0
a[640] = 0
a[641] = 0
a[642] = 0
a[643] = 0
a[644] = 0
a[645] = 0
a[646] = 0
a[647] = 0
a[648] = 0
a[649] = 0
a[650] = 0
a[651] = 0
a[652] = 0
a[653] = 0
a[654] = 0
a[655] = 0
a[656] = 0
a[657] = 0
a[658] = 0
a[659] = 0
a[660] = 0
a[661] = 0
a[662] = 0
a[663] = 0
a[664] = 0
a[665] = 0
a[666] = 0
a[667] = 0
a[668] = 0
a[669] = 0
a[670] = 0
a[671] = 0
a[672] = 0
a[673] = 0
a[674] = 0
a[675] = 0
a[676] = 0
a[677] = 0
a[678] = 0
a[679] = 0
a[680] = 0
a[681] = 0
a[682] = 0
a[683] = 0
a[684] = 0
a[685] = 0
a[686] = 0
a[687] = 0
a[688] = 0
a[689] = 0
a[690] = 0
a[691] = 0
a[692] = 0
a[693] = 0
a[694] = 0
a[695] = 0
a[696] = 0
a[697] = 0
a[698] = 0
a[699] = 0
a[700] = 0
a[701] = 0
a[702] = 0
a[703] = 0
a[704] = 0
a[705] = 0
a[706] = 0
a[707] = 0
a[708] = 0
a[709] = 0
a[710] = 0
a[711] = 0
a[712] = 0
a[713] = 0
a[714] = 0
a[715] = 0
a[716] = 0
a[717] = 0
a[718] = 0
a[719] = 0
a[720] = 0
a[721] = 0
a[722] = 0
a[723] = 0
a[724] = 0
a[725] = 0
a[726] = 0
a[727] = 0
a[728] = 0
a[729] = 0
a[730] = 0
a[731] = 0
a[732] = 0
a[733] = 0
a[734] = 0
a[735] = 0
a[736] = 0
a[737] = 0
a[738] = 0
a[739] = 0
a[740] = 0
a[741] = 0
a[742] = 0
a[743] = 0
a[744] = 0
a[745] = 0
a[746] = 0
a[747] = 0
a[748] = 0
a[749] = 0
a[750] = 0
a[751] = 0
a[752] = 0
a[753] = 0
a[754] = 0
a[755] = 0
a[756] = 0
a[757] = 0
a[758] = 0
a[759] = 0
a[760] = 0
a[761] = 0
a[762] = 0
a[763] = 0
a[764] = 0
a[765] = 0
a[766] = 0
a[767] = 0
a[768] = 0
a[769] = 0
a[770] = 0
a[771] = 0
a[772] = 0
a[773] = 0
a[774] = 0
a[775] = 0
a[776] = 0
a[777] = 0
a[778] = 0
a[779] = 0
a[780] = 0
a[781] = 0
a[782] = 0
a[783] = 0
a[784] = 0
a[785] = 0
a[786] = 0
a[787] = 0
a[788] = 0
a[789] = 0
a[790] = 0
a[791] = 0
a[792] = 0
a[793] = 0
a[794] = 0
a[795] = 0
a[796] = 0
a[797] = 0
a[798] = 0
a[799] = 0
a[800] = 0
a[801] = 0
a[802] = 0
a[803] = 0
a[804] = 0
a[805] = 0
a[806] = 0
a[807] = 0
a[808] = 0
a[809] = 0
a[810] = 0
a[811] = 0
a[812] = 0
a[813] = 0
a[814] = 0
a[815] = 0
a[816] = 0
a[817] = 0
a[818] = 0
a[819] = 0
a[820] = 0
a[821] = 0
a[822] = 0
a[823] = 0
a[824] = 0
a[825] = 0
a[826] = 0
a[827] = 0
a[828] = 0
a[829] = 0
a[830] = 0
a[831] = 0
a[832] = 0
a[833] = 0
a[834] = 0
a[835] = 0
a[836] = 0
a[837] = 0
a[838] = 0
a[839] = 0
a[840] = 0
a[841] = 0
a[842] = 0
a[843] = 0
a[844] = 0
a[845] = 0
a[846] = 0
a[847] = 0
a[848] = 0
a[849] = 0
a[850] = 0
a[851] = 0
a[852] = 0
a[853] = 0
a[854] = 0
a[855] = 0
a[856] = 0
a[857] = 0
a[858] = 0
a[859] = 0
a[860] = 0
a[861] = 0
a[862] = 0
a[863] = 0
a[864] = 0
a[865] = 0
a[866] = 0
a[867] = 0
a[868] = 0
a[869] = 0
a[870] = 0
a[871] = 0
a[872] = 0
a[873] = 0
a[874] = 0
a[875] = 0
a[876] = 0
a[877] = 0
a[878] = 0
a[879] = 0
a[880] = 0
a[881] = 0
a[882] = 0
a[883] = 0
a[884] = 0
a[885] = 0
a[886] = 0
a[887] = 0
a[888] = 0
a[889] = 0
a[890] = 0
a[891] = 0
a[892] = 0
a[893] = 0
a[894] = 0
a[895] = 0
a[896] = 0
a[897] = 0
a[898] = 0
a[899] = 0
a[900] = 0
a[901] = 0
a[902] = 0
a[903] = 0
a[904] = 0
a[905] = 0
a[906] = 0
a[907] = 0
a[908] = 0
a[909] = 0
a[910] = 0
a[911] = 0
a[912] = 0
a[913] = 0
a[914] = 0
a[915] = 0
a[916] = 0
a[917] = 0
a[918] = 0
a[919] = 0
a[920] = 0
a[921] = 0
a[922] = 0
a[923] = 0
a[924] = 0
a[925] = 0
a[926] = 0
a[927] = 0
a[928] = 0
a[929] = 0
a[930] = 0
a[931] = 0
a[932] = 0
a[933] = 0
a[934] = 0
a[935] = 0
a[936] = 0
a[937] = 0
a[938] = 0
a[939] = 0
a[940] = 0
a[941] = 0
a[942] = 0
a[943] = 0
a[944] = 0
a[945] = 0
a[946] = 0
a[947] = 0
a[948] = 0
a[949] = 0
a[950] = 0
a[951] = 0
a[952] = 0
a[953] = 0
a[954] = 0
a[955] = 0
a[956] = 0
a[957] = 0
a[958] = 0
a[959] = 0
a[960] = 0
a[961] = 0
a[962] = 0
a[963] = 0
a[964] = 0
a[965] = 0
a[966] = 0
a[967] = 0
a[968] = 0
a[969] = 0
a[970] = 0
a[971] = 0
a[972] = 0
a[973] = 0
a[974] = 0
a[975] = 0
a[976] = 0
a[977] = 0
a[978] = 0
a[979] = 0
a[980] = 0
a[981] = 0
a[982] = 0
a[983] = 0
a[984] = 0
a[985] = 0
a[986] = 0
a[987] = 0
a[988] = 0
a[989] = 0
a[990] = 0
a[991] = 0
a[992] = 0
a[993] = 0
a[994] = 0
a[995] = 0
a[996] = 0
a[997] = 0
a[998] = 0
a[999] = 0
[dev.typeparams] cmd/compile/internal/inline: refactor mkinlcall This CL refactors mkinlcall by extracting the core InlinedCallExpr construction code into a new "oldInline" function, and adds a new "NewInline" hook point that can be overriden with a new inliner implementation that only needs to worry about the details of constructing the InlinedCallExpr. It also moves the delayretvars optimization check into CanInline, so it's performed just once per inlinable function rather than once for each inlined call. Finally, it skips printing the function body about to be inlined (and updates the couple of regress tests that expected this output). We already report the inline body as it was saved, and this diagnostic is only applicable to the current inliner, which clones existing function body instances. In the unified IR inliner, we'll directly construct inline bodies from the serialized representation. Change-Id: Ibdbe617da83c07665dcbda402cc8d4d4431dde2f Reviewed-on: https://go-review.googlesource.com/c/go/+/323290 Trust: Matthew Dempsky <mdempsky@google.com> Trust: Dan Scales <danscales@google.com> Run-TryBot: Matthew Dempsky <mdempsky@google.com> TryBot-Result: Go Bot <gobot@golang.org> Reviewed-by: Dan Scales <danscales@google.com>
2021-05-27 03:47:04 -06:00
x := small(a) // ERROR "inlining call to small"
cmd/compile: set stricter inlining threshold in large functions If we're compiling a large function, be more picky about how big the function we're inlining is. If the function is >5000 nodes, we lower the inlining threshold from a cost of 80 to 20. Turns out reflect.Value's cost is exactly 80. That's the function at issue in #26546. 20 was chosen as a proxy for "inlined body is smaller than the call would be". Simple functions still get inlined, like this one at cost 7: func ifaceIndir(t *rtype) bool { return t.kind&kindDirectIface == 0 } 5000 nodes was chosen as the big function size. Here are all the 5000+ node (~~1000+ lines) functions in the stdlib: 5187 cmd/internal/obj/arm (*ctxt5).asmout 6879 cmd/internal/obj/s390x (*ctxtz).asmout 6567 cmd/internal/obj/ppc64 (*ctxt9).asmout 9643 cmd/internal/obj/arm64 (*ctxt7).asmout 5042 cmd/internal/obj/x86 (*AsmBuf).doasm 8768 cmd/compile/internal/ssa rewriteBlockAMD64 8878 cmd/compile/internal/ssa rewriteBlockARM 8344 cmd/compile/internal/ssa rewriteValueARM64_OpARM64OR_20 7916 cmd/compile/internal/ssa rewriteValueARM64_OpARM64OR_30 5427 cmd/compile/internal/ssa rewriteBlockARM64 5126 cmd/compile/internal/ssa rewriteValuePPC64_OpPPC64OR_50 6152 cmd/compile/internal/ssa rewriteValuePPC64_OpPPC64OR_60 6412 cmd/compile/internal/ssa rewriteValuePPC64_OpPPC64OR_70 6486 cmd/compile/internal/ssa rewriteValuePPC64_OpPPC64OR_80 6534 cmd/compile/internal/ssa rewriteValuePPC64_OpPPC64OR_90 6534 cmd/compile/internal/ssa rewriteValuePPC64_OpPPC64OR_100 6534 cmd/compile/internal/ssa rewriteValuePPC64_OpPPC64OR_110 6675 cmd/compile/internal/gc typecheck1 5433 cmd/compile/internal/gc walkexpr 14070 cmd/vendor/golang.org/x/arch/arm64/arm64asm decodeArg There are a lot more smaller (~1000 node) functions in the stdlib. The function in #26546 has 12477 nodes. At some point it might be nice to have a better heuristic for "inlined body is smaller than the call", a non-cliff way to scale down the cost as the function gets bigger, doing cheaper inlined calls first, etc. All that can wait for another release. I'd like to do this CL for 1.11. Fixes #26546 Update #17566 Change-Id: Idda13020e46ec2b28d79a17217f44b189f8139ac Reviewed-on: https://go-review.googlesource.com/125516 Run-TryBot: Keith Randall <khr@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: David Chase <drchase@google.com>
2018-07-23 14:09:48 -06:00
y := medium(a) // The crux of this test: medium is not inlined.
return x + y
}
Copy Permalink