1Dalvik "mterp" README 2 3NOTE: Find rebuilding instructions at the bottom of this file. 4 5 6==== Overview ==== 7 8This is the source code for the Dalvik interpreter. The core of the 9original version was implemented as a single C function, but to improve 10performance we rewrote it in assembly. To make this and future assembly 11ports easier and less error-prone, we used a modular approach that allows 12development of platform-specific code one opcode at a time. 13 14The original all-in-one-function C version still exists as the "portable" 15interpreter, and is generated using the same sources and tools that 16generate the platform-specific versions. One form of the portable 17interpreter includes support for profiling and debugging features, and 18is included even if we have a platform-optimized implementation. 19 20Every configuration has a "config-*" file that controls how the sources 21are generated. The sources are written into the "out" directory, where 22they are picked up by the Android build system. 23 24The best way to become familiar with the interpreter is to look at the 25generated files in the "out" directory, such as out/InterpC-portstd.c, 26rather than trying to look at the various component pieces in (say) 27armv5te. 28 29 30==== Platform-specific source generation ==== 31 32The architecture-specific config files determine what goes into two 33generated output files (InterpC-<arch>.c, InterpAsm-<arch>.S). The goal is 34to make it easy to swap C and assembly sources during initial development 35and testing, and to provide a way to use architecture-specific versions of 36some operations (e.g. making use of PLD instructions on ARMv6 or avoiding 37CLZ on ARMv4T). 38 39Two basic assumptions are made about the operation of the interpreter: 40 41 - The assembly version uses fixed-size areas for each instruction 42 (e.g. 64 bytes). "Overflow" code is tacked on to the end. 43 - When a C implementation is desired, the assembly version packs all 44 local state into a "glue" struct, and passes that into the C function. 45 Updates to the state are pulled out of the "glue" on return. 46 47The "arch" value should indicate an architecture family with common 48programming characteristics, so "armv5te" would work for all ARMv5TE CPUs, 49but might not be backward- or forward-compatible. (We *might* want to 50specify the ABI model as well, e.g. "armv5te-eabi", but currently that adds 51verbosity without value.) 52 53 54==== Config file format ==== 55 56The config files are parsed from top to bottom. Each line in the file 57may be blank, hold a comment (line starts with '#'), or be a command. 58 59The commands are: 60 61 handler-size <bytes> 62 63 Specify the size of the assembly region, in bytes. On most platforms 64 this will need to be a power of 2. 65 66 import <filename> 67 68 The specified file is included immediately, in its entirety. No 69 substitutions are performed. ".c" and ".h" files are copied to the 70 C output, ".S" files are copied to the asm output. 71 72 asm-stub <filename> 73 74 The named file will be included whenever an assembly "stub" is needed. 75 Text substitution is performed on the opcode name. 76 77 op-start <directory> 78 79 Indicates the start of the opcode list. Must precede any "op" 80 commands. The specified directory is the default location to pull 81 instruction files from. 82 83 op <opcode> <directory> 84 85 Can only appear after "op-start" and before "op-end". Overrides the 86 default source file location of the specified opcode. The opcode 87 definition will come from the specified file, e.g. "op OP_NOP armv5te" 88 will load from "armv5te/OP_NOP.S". A substitution dictionary will be 89 applied (see below). 90 91 op-end 92 93 Indicates the end of the opcode list. All 256 opcodes are emitted 94 when this is seen, followed by any code that didn't fit inside the 95 fixed-size instruction handler space. 96 97 98The order of "op" directives is not significant; the generation tool will 99extract ordering info from the VM sources. 100 101Typically the form in which most opcodes currently exist is used in 102the "op-start" directive. For a new port you would start with "c", 103and add architecture-specific "op" entries as you write instructions. 104When complete it will default to the target architecture, and you insert 105"c" ops to stub out platform-specific code. 106 107For the <directory> specified in the "op" command, the "c" directory 108is special in two ways: (1) the sources are assumed to be C code, and 109will be inserted into the generated C file; (2) when a C implementation 110is emitted, a "glue stub" is emitted in the assembly source file. 111(The generator script always emits 256 assembly instructions, unless 112"asm-stub" was left blank, in which case it only emits some labels.) 113 114 115==== Instruction file format ==== 116 117The assembly instruction files are simply fragments of assembly sources. 118The starting label will be provided by the generation tool, as will 119declarations for the segment type and alignment. The expected target 120assembler is GNU "as", but others will work (may require fiddling with 121some of the pseudo-ops emitted by the generation tool). 122 123The C files do a bunch of fancy things with macros in an attempt to share 124code with the portable interpreter. (This is expected to be reduced in 125the future.) 126 127A substitution dictionary is applied to all opcode fragments as they are 128appended to the output. Substitutions can look like "$value" or "${value}". 129 130The dictionary always includes: 131 132 $opcode - opcode name, e.g. "OP_NOP" 133 $opnum - opcode number, e.g. 0 for OP_NOP 134 $handler_size_bytes - max size of an instruction handler, in bytes 135 $handler_size_bits - max size of an instruction handler, log 2 136 137Both C and assembly sources will be passed through the C pre-processor, 138so you can take advantage of C-style comments and preprocessor directives 139like "#define". 140 141Some generator operations are available. 142 143 %include "filename" [subst-dict] 144 145 Includes the file, which should look like "armv5te/OP_NOP.S". You can 146 specify values for the substitution dictionary, using standard Python 147 syntax. For example, this: 148 %include "armv5te/unop.S" {"result":"r1"} 149 would insert "armv5te/unop.S" at the current file position, replacing 150 occurrences of "$result" with "r1". 151 152 %default <subst-dict> 153 154 Specify default substitution dictionary values, using standard Python 155 syntax. Useful if you want to have a "base" version and variants. 156 157 %break 158 159 Identifies the split between the main portion of the instruction 160 handler (which must fit in "handler-size" bytes) and the "sister" 161 code, which is appended to the end of the instruction handler block. 162 163 %verify "message" 164 165 Leave a note to yourself about what needs to be tested. (This may 166 turn into something more interesting someday; for now, it just gets 167 stripped out before the output is generated.) 168 169The generation tool does *not* print a warning if your instructions 170exceed "handler-size", but the VM will abort on startup if it detects an 171oversized handler. On architectures with fixed-width instructions this 172is easy to work with, on others this you will need to count bytes. 173 174 175==== Using C constants from assembly sources ==== 176 177The file "common/asm-constants.h" has some definitions for constant 178values, structure sizes, and struct member offsets. The format is fairly 179restricted, as simple macros are used to massage it for use with both C 180(where it is verified) and assembly (where the definitions are used). 181 182If a constant in the file becomes out of sync, the VM will log an error 183message and abort during startup. 184 185 186==== Development tips ==== 187 188If you need to debug the initial piece of an opcode handler, and your 189debug code expands it beyond the handler size limit, you can insert a 190generic header at the top: 191 192 b ${opcode}_start 193%break 194${opcode}_start: 195 196If you already have a %break, it's okay to leave it in place -- the second 197%break is ignored. 198 199 200==== Rebuilding ==== 201 202If you change any of the source file fragments, you need to rebuild the 203combined source files in the "out" directory. Make sure the files in 204"out" are editable, then: 205 206 $ cd mterp 207 $ ./rebuild.sh 208 209As of this writing, this requires Python 2.5. You may see inscrutible 210error messages or just general failure if you have a different version 211of Python installed. 212 213The ultimate goal is to have the build system generate the necessary 214output files without requiring this separate step, but we're not yet 215ready to require Python in the build. 216