github.com/tinygo-org/tinygo@v0.31.3-0.20240404173401-90b0bf646c27/interp/README.md (about) 1 # Partial evaluation of initialization code in Go 2 3 For several reasons related to code size and memory consumption (see below), it 4 is best to try to evaluate as much initialization code at compile time as 5 possible and only run unknown expressions (e.g. external calls) at runtime. This 6 is in practice a partial evaluator of the `runtime.initAll` function, which 7 calls each package initializer. 8 9 This package is a rewrite of a previous partial evaluator that worked 10 directly on LLVM IR and used the module and LLVM constants as intermediate 11 values. This newer version instead uses a mostly Go intermediate form. It 12 compiles functions and extracts relevant data first (compiler.go), then 13 executes those functions (interpreter.go) in a memory space that can be 14 rolled back per function (memory.go). This means that it is not necessary to 15 scan functions to see whether they can be run at compile time, which was very 16 error prone. Instead it just tries to execute everything and if it hits 17 something it cannot interpret (such as a store to memory-mapped I/O) it rolls 18 back the execution of that function and runs the function at runtime instead. 19 All in all, this design provides several benefits: 20 21 * Much better error handling. By being able to revert to runtime execution 22 without the need for scanning functions, this version is able to 23 automatically work around many bugs in the previous implementation. 24 * More correct memory model. This is not inherent to the new design, but the 25 new design also made the memory model easier to reason about. 26 * Faster execution of initialization code. While it is not much faster for 27 normal interpretation (maybe 25% or so) due to the compilation overhead, 28 it should be a whole lot faster for loops as it doesn't have to call into 29 LLVM (via CGo) for every operation. 30 31 As mentioned, this partial evaulator comes in three parts: a compiler, an 32 interpreter, and a memory manager. 33 34 ## Compiler 35 36 The main task of the compiler is that it extracts all necessary data from 37 every instruction in a function so that when this instruction is interpreted, 38 no additional CGo calls are necessary. This is not currently done for all 39 instructions (`runtime.alloc` is a notable exception), but at least it does 40 so for the vast majority of instructions. 41 42 ## Interpreter 43 44 The interpreter runs an instruction just like it would if it were executed 45 'for real'. The vast majority of instructions can be executed at compile 46 time. As indicated above, some instructions need to be executed at runtime 47 instead. 48 49 ## Memory 50 51 Memory is represented as objects (the `object` type) that contains data that 52 will eventually be stored in a global and values (the `value` interface) that 53 can be worked with while running the interpreter. Values therefore are only 54 used locally and are always passed by value (just like most LLVM constants) 55 while objects represent the backing storage (like LLVM globals). Some values 56 are pointer values, and point to an object. 57 58 Importantly, this partial evaluator can roll back the execution of a 59 function. This is implemented by creating a new memory view per function 60 activation, which makes sure that any change to a global (such as a store 61 instruction) is stored in the memory view. It creates a copy of the object 62 and stores that in the memory view to be modified. Once the function has 63 executed successfully, all these modified objects are then copied into the 64 parent function, up to the root function invocation which (on successful 65 execution) writes the values back into the LLVM module. This way, function 66 invocations can be rolled back without leaving a trace. 67 68 Pointer values point to memory objects, but not to a particular memory 69 object. Every memory object is given an index, and pointers use that index to 70 look up the current active object for the pointer to load from or to copy 71 when storing to it. 72 73 Rolling back a function should roll back everything, including the few 74 instructions emitted at runtime. This is done by treating instructions much 75 like memory objects and removing the created instructions when necessary. 76 77 ## Why is this necessary? 78 79 A partial evaluator is hard to get right, so why go through all the trouble of 80 writing one? 81 82 The answer is that globals with initializers are much easier to optimize by 83 LLVM than initialization code. Also, there are a few other benefits: 84 85 * Dead globals are trivial to optimize away. 86 * Constant globals are easier to detect. Remember that Go does not have global 87 constants in the same sense as that C has them. Constants are useful because 88 they can be propagated and provide some opportunities for other 89 optimizations (like dead code elimination when branching on the contents of 90 a global). 91 * Constants are much more efficient on microcontrollers, as they can be 92 allocated in flash instead of RAM. 93 94 The Go SSA package does not create constant initializers for globals. 95 Instead, it emits initialization functions, so if you write the following: 96 97 ```go 98 var foo = []byte{1, 2, 3, 4} 99 ``` 100 101 It would generate something like this: 102 103 ```go 104 var foo []byte 105 106 func init() { 107 foo = make([]byte, 4) 108 foo[0] = 1 109 foo[1] = 2 110 foo[2] = 3 111 foo[3] = 4 112 } 113 ``` 114 115 This is of course hugely wasteful, it's much better to create `foo` as a 116 global array instead of initializing it at runtime. 117 118 For more details, see [this section of the 119 documentation](https://tinygo.org/compiler-internals/differences-from-go/).