1 // Copyright 2018 The Go Authors. All rights reserved.
2 // Use of this source code is governed by a BSD-style
3 // license that can be found in the LICENSE file.
4
5 The codegen directory contains code generation tests for the gc
6 compiler.
7
8
9 - Introduction
10
11 The test harness compiles Go code inside files in this directory and
12 matches the generated assembly (the output of `go tool compile -S`)
13 against a set of regexps to be specified in comments that follow a
14 special syntax (described below). The test driver is implemented as a
15 step of the top-level test/run.go suite, called "asmcheck".
16
17 The codegen harness is part of the all.bash test suite, but for
18 performance reasons only the codegen tests for the host machine's
19 GOARCH are enabled by default, and only on GOOS=linux.
20
21 To perform comprehensive tests for all the supported architectures
22 (even on a non-Linux system), one can run the following command
23
24 $ ../bin/go run run.go -all_codegen -v codegen
25
26 in the top-level test directory. This is recommended after any change
27 that affect the compiler's code.
28
29 The test harness compiles the tests with the same go toolchain that is
30 used to run run.go. After writing tests for a newly added codegen
31 transformation, it can be useful to first run the test harness with a
32 toolchain from a released Go version (and verify that the new tests
33 fail), and then re-runnig the tests using the devel toolchain.
34
35
36 - Regexps comments syntax
37
38 Instructions to match are specified inside plain comments that start
39 with an architecture tag, followed by a colon and a quoted Go-style
40 regexp to be matched. For example, the following test:
41
42 func Sqrt(x float64) float64 {
43 // amd64:"SQRTSD"
44 // arm64:"FSQRTD"
45 return math.Sqrt(x)
46 }
47
48 verifies that math.Sqrt calls are intrinsified to a SQRTSD instruction
49 on amd64, and to a FSQRTD instruction on arm64.
50
51 It is possible to put multiple architectures checks into the same
52 line, as:
53
54 // amd64:"SQRTSD" arm64:"FSQRTD"
55
56 although this form should be avoided when doing so would make the
57 regexps line excessively long and difficult to read.
58
59 Comments that are on their own line will be matched against the first
60 subsequent non-comment line. Inline comments are also supported; the
61 regexp will be matched against the code found on the same line:
62
63 func Sqrt(x float64) float64 {
64 return math.Sqrt(x) // arm:"SQRTD"
65 }
66
67 It's possible to specify a comma-separated list of regexps to be
68 matched. For example, the following test:
69
70 func TZ8(n uint8) int {
71 // amd64:"BSFQ","ORQ\t\\$256"
72 return bits.TrailingZeros8(n)
73 }
74
75 verifies that the code generated for a bits.TrailingZeros8 call on
76 amd64 contains both a "BSFQ" instruction and an "ORQ $256".
77
78 Note how the ORQ regex includes a tab char (\t). In the Go assembly
79 syntax, operands are separated from opcodes by a tabulation.
80
81 Regexps can be quoted using either " or `. Special characters must be
82 escaped accordingly. Both of these are accepted, and equivalent:
83
84 // amd64:"ADDQ\t\\$3"
85 // amd64:`ADDQ\t\$3`
86
87 and they'll match this assembly line:
88
89 ADDQ $3
90
91 Negative matches can be specified using a - before the quoted regexp.
92 For example:
93
94 func MoveSmall() {
95 x := [...]byte{1, 2, 3, 4, 5, 6, 7}
96 copy(x[1:], x[:]) // arm64:-".*memmove"
97 }
98
99 verifies that NO memmove call is present in the assembly generated for
100 the copy() line.
101
102
103 - Architecture specifiers
104
105 There are three different ways to specify on which architecture a test
106 should be run:
107
108 * Specify only the architecture (eg: "amd64"). This indicates that the
109 check should be run on all the supported architecture variants. For
110 instance, arm checks will be run against all supported GOARM
111 variations (5,6,7).
112 * Specify both the architecture and a variant, separated by a slash
113 (eg: "arm/7"). This means that the check will be run only on that
114 specific variant.
115 * Specify the operating system, the architecture and the variant,
116 separated by slashes (eg: "plan9/386/sse2", "plan9/amd64/"). This is
117 needed in the rare case that you need to do a codegen test affected
118 by a specific operating system; by default, tests are compiled only
119 targeting linux.
120
121
122 - Remarks, and Caveats
123
124 -- Write small test functions
125
126 As a general guideline, test functions should be small, to avoid
127 possible interactions between unrelated lines of code that may be
128 introduced, for example, by the compiler's optimization passes.
129
130 Any given line of Go code could get assigned more instructions than it
131 may appear from reading the source. In particular, matching all MOV
132 instructions should be avoided; the compiler may add them for
133 unrelated reasons and this may render the test ineffective.
134
135 -- Line matching logic
136
137 Regexps are always matched from the start of the instructions line.
138 This means, for example, that the "MULQ" regexp is equivalent to
139 "^MULQ" (^ representing the start of the line), and it will NOT match
140 the following assembly line:
141
142 IMULQ $99, AX
143
144 To force a match at any point of the line, ".*MULQ" should be used.
145
146 For the same reason, a negative regexp like -"memmove" is not enough
147 to make sure that no memmove call is included in the assembly. A
148 memmove call looks like this:
149
150 CALL runtime.memmove(SB)
151
152 To make sure that the "memmove" symbol does not appear anywhere in the
153 assembly, the negative regexp to be used is -".*memmove".
154
View as plain text