Revision c22889e76cb9b7fd8a4710d9bf53e827aaa907e4 authored by Shuhei Kadowaki on 24 August 2021, 04:21:29 UTC, committed by Shuhei Kadowaki on 26 October 2021, 14:45:33 UTC
Currently our constant-prop' heuristics work in the following way:
1. `const_prop_entry_heuristic`
2. `const_prop_argument_heuristic` & `const_prop_rettype_heuristic`
3. `force_const_prop` custom heuristic & `!const_prop_function_heuristic`
4. `MethodInstance` specialization and `const_prop_methodinstance_heuristic`

This PR changes it so that the step 1. now works like:

1. `force_const_prop` custom heuristic & `const_prop_entry_heuristic`

and the steps 2., 3. and 4. don't change

This change particularly allows us to more forcibly constant-propagate
for `getproperty` and `setproperty!`, and inline them more, e.g.:
```julia
mutable struct Foo
    val
    _::Int
end

function setter(xs)
    for x in xs
        x.val = nothing # `setproperty!` can be inlined with this PR
    end
end
```

It might be useful because now we can intervene into the constant-prop'
heuristic in a more reliable way with the `aggressive_constprop` interface.

I did the simple benchmark below, and it looks like this change doesn't
cause the latency problem for this particular example:
```zsh
~/julia master aviatesk@amdci2 6s
❯ ./usr/bin/julia -e '@time using Plots; @time plot(rand(10,3))'
  3.708500 seconds (7.28 M allocations: 506.128 MiB, 3.45% gc time, 1.13% compilation time)
  2.817794 seconds (3.45 M allocations: 195.127 MiB, 7.84% gc time, 53.76% compilation time)

~/julia avi/forceconstantprop aviatesk@amdci2 6s
❯ ./usr/bin/julia -e '@time using Plots; @time plot(rand(10,3))'
  3.622109 seconds (7.02 M allocations: 481.710 MiB, 4.19% gc time, 1.17% compilation time)
  2.863419 seconds (3.44 M allocations: 194.210 MiB, 8.02% gc time, 53.53% compilation time)
```
1 parent da71d29
Raw File
codegen_shared.h
// This file is a part of Julia. License is MIT: https://julialang.org/license

#include <utility>
#include <llvm/ADT/ArrayRef.h>
#include <llvm/Support/Debug.h>
#include <llvm/IR/DebugLoc.h>
#include <llvm/IR/IRBuilder.h>

enum AddressSpace {
    Generic = 0,
    Tracked = 10,
    Derived = 11,
    CalleeRooted = 12,
    Loaded = 13,
    FirstSpecial = Tracked,
    LastSpecial = Loaded,
};

// JLCALL with API arguments ([extra], arg0, arg1, arg2, ...) has the following ABI calling conventions defined:
#define JLCALL_F_CC (CallingConv::ID)37     // (jl_value_t *arg0, jl_value_t **argv, uint32_t nargv)
#define JLCALL_F2_CC (CallingConv::ID)38    // (jl_value_t *arg0, jl_value_t **argv, uint32_t nargv, jl_value_t *extra)

// return how many Tracked pointers are in T (count > 0),
// and if there is anything else in T (all == false)
struct CountTrackedPointers {
    unsigned count = 0;
    bool all = true;
    bool derived = false;
    CountTrackedPointers(llvm::Type *T);
};

unsigned TrackWithShadow(llvm::Value *Src, llvm::Type *T, bool isptr, llvm::Value *Dst, llvm::IRBuilder<> &irbuilder);
std::vector<llvm::Value*> ExtractTrackedValues(llvm::Value *Src, llvm::Type *STy, bool isptr, llvm::IRBuilder<> &irbuilder, llvm::ArrayRef<unsigned> perm_offsets={});

static inline void llvm_dump(llvm::Value *v)
{
    v->print(llvm::dbgs(), true);
    llvm::dbgs() << "\n";
}

static inline void llvm_dump(llvm::Type *v)
{
    v->print(llvm::dbgs(), true);
    llvm::dbgs() << "\n";
}

static inline void llvm_dump(llvm::Function *f)
{
    f->print(llvm::dbgs(), nullptr, false, true);
}

static inline void llvm_dump(llvm::Module *m)
{
    m->print(llvm::dbgs(), nullptr);
}

static inline void llvm_dump(llvm::Metadata *m)
{
    m->print(llvm::dbgs());
    llvm::dbgs() << "\n";
}

static inline void llvm_dump(llvm::DebugLoc *dbg)
{
    dbg->print(llvm::dbgs());
    llvm::dbgs() << "\n";
}
back to top