Switch LLVM codegen of Ptr{T} to an actual pointer type. (#53687)

This PR switches our code generation for `Ptr{T}` from `i64` to an
actual LLVM pointer type (`ptr` when using opaque pointers, an untyped
`i8*` otherwise). The main motivation is to simplify `llvmcall` usage
(doing away with the `inttoptr`/`ptrtoint` conversions), and also make
it possible to simply use `ccall` to call intrinsics with `Ptr`-valued
arguments (where we currently always need `llvmcall` for converting to
an actual pointer).

Changing codegen like this is a breaking change for `llvmcall` users,
but I've added backwards compatibility and a deprecation warning.

Before:

```llvm
julia> @code_llvm pointer([])
define i64 @julia_pointer_1542(ptr noundef nonnull align 8 dereferenceable(24) %"x::Array") #0 {
top:
; ┌ @ pointer.jl:65 within `cconvert`
   %0 = load ptr, ptr %"x::Array", align 8
; └
; ┌ @ pointer.jl:90 within `unsafe_convert`
; │┌ @ pointer.jl:30 within `convert`
    %bitcast_coercion = ptrtoint ptr %0 to i64
    ret i64 %bitcast_coercion
; └└
}
```

After:

```llvm
julia> @code_llvm pointer([])
define ptr @julia_pointer_3880(ptr noundef nonnull align 8 dereferenceable(24) %"x::Array") #0 {
top:
; ┌ @ pointer.jl:65 within `cconvert`
   %0 = load ptr, ptr %"x::Array", align 8
; └
; ┌ @ pointer.jl:90 within `unsafe_convert`
; │┌ @ pointer.jl:30 within `convert`
    ret ptr %0
; └└
}
```

This also simplifies "real code", e.g., when `ccall` converts an Array
to a pointer, resulting in some more optimization opportunities.

Author: maleadt

NEWS.md

@@ -26,6 +26,11 @@ Language changes
 Compiler/Runtime improvements
 -----------------------------
 
+- Generated LLVM IR now uses actual pointer types instead of passing pointers as integers.
+  This affects `llvmcall`: Inline LLVM IR should be updated to use `i8*` or `ptr` instead of
+  `i32` or `i64`, and remove unneeded `ptrtoint`/`inttoptr` conversions. For compatibility,
+  IR with integer pointers is still supported, but generates a deprecation warning. ([#53687])
+
 Command-line option changes
 ---------------------------
 

base/atomics.jl

@@ -364,13 +364,13 @@ for typ in atomictypes
     irt = "$ilt, $ilt*"
     @eval getindex(x::Atomic{$typ}) =
         GC.@preserve x llvmcall($"""
-                 %ptr = inttoptr i$WORD_SIZE %0 to $lt*
+                 %ptr = bitcast i8* %0 to $lt*
                  %rv = load atomic $rt %ptr acquire, align $(gc_alignment(typ))
                  ret $lt %rv
                  """, $typ, Tuple{Ptr{$typ}}, unsafe_convert(Ptr{$typ}, x))
     @eval setindex!(x::Atomic{$typ}, v::$typ) =
         GC.@preserve x llvmcall($"""
-                 %ptr = inttoptr i$WORD_SIZE %0 to $lt*
+                 %ptr = bitcast i8* %0 to $lt*
                  store atomic $lt %1, $lt* %ptr release, align $(gc_alignment(typ))
                  ret void
                  """, Cvoid, Tuple{Ptr{$typ}, $typ}, unsafe_convert(Ptr{$typ}, x), v)
@@ -379,7 +379,7 @@ for typ in atomictypes
     if typ <: Integer
         @eval atomic_cas!(x::Atomic{$typ}, cmp::$typ, new::$typ) =
             GC.@preserve x llvmcall($"""
-                     %ptr = inttoptr i$WORD_SIZE %0 to $lt*
+                     %ptr = bitcast i8* %0 to $lt*
                      %rs = cmpxchg $lt* %ptr, $lt %1, $lt %2 acq_rel acquire
                      %rv = extractvalue { $lt, i1 } %rs, 0
                      ret $lt %rv
@@ -388,7 +388,7 @@ for typ in atomictypes
     else
         @eval atomic_cas!(x::Atomic{$typ}, cmp::$typ, new::$typ) =
             GC.@preserve x llvmcall($"""
-                     %iptr = inttoptr i$WORD_SIZE %0 to $ilt*
+                     %iptr = bitcast i8* %0 to $ilt*
                      %icmp = bitcast $lt %1 to $ilt
                      %inew = bitcast $lt %2 to $ilt
                      %irs = cmpxchg $ilt* %iptr, $ilt %icmp, $ilt %inew acq_rel acquire
@@ -411,15 +411,15 @@ for typ in atomictypes
         if typ <: Integer
             @eval $fn(x::Atomic{$typ}, v::$typ) =
                 GC.@preserve x llvmcall($"""
-                         %ptr = inttoptr i$WORD_SIZE %0 to $lt*
+                         %ptr = bitcast i8* %0 to $lt*
                          %rv = atomicrmw $rmw $lt* %ptr, $lt %1 acq_rel
                          ret $lt %rv
                          """, $typ, Tuple{Ptr{$typ}, $typ}, unsafe_convert(Ptr{$typ}, x), v)
         else
             rmwop === :xchg || continue
             @eval $fn(x::Atomic{$typ}, v::$typ) =
                 GC.@preserve x llvmcall($"""
-                         %iptr = inttoptr i$WORD_SIZE %0 to $ilt*
+                         %iptr = bitcast i8* %0 to $ilt*
                          %ival = bitcast $lt %1 to $ilt
                          %irv = atomicrmw $rmw $ilt* %iptr, $ilt %ival acq_rel
                          %rv = bitcast $ilt %irv to $lt

base/compiler/tfuncs.jl

@@ -184,8 +184,6 @@ add_tfunc(sdiv_int, 2, 2, math_tfunc, 20)
 add_tfunc(udiv_int, 2, 2, math_tfunc, 20)
 add_tfunc(srem_int, 2, 2, math_tfunc, 20)
 add_tfunc(urem_int, 2, 2, math_tfunc, 20)
-add_tfunc(add_ptr, 2, 2, math_tfunc, 1)
-add_tfunc(sub_ptr, 2, 2, math_tfunc, 1)
 add_tfunc(neg_float, 1, 1, math_tfunc, 1)
 add_tfunc(add_float, 2, 2, math_tfunc, 2)
 add_tfunc(sub_float, 2, 2, math_tfunc, 2)
@@ -662,6 +660,9 @@ function pointer_eltype(@nospecialize(ptr))
     return Any
 end
 
+@nospecs function pointerarith_tfunc(𝕃::AbstractLattice, ptr, offset)
+    return ptr
+end
 @nospecs function pointerref_tfunc(𝕃::AbstractLattice, a, i, align)
     return pointer_eltype(a)
 end
@@ -705,6 +706,8 @@ end
     end
     return ccall(:jl_apply_cmpswap_type, Any, (Any,), T) where T
 end
+add_tfunc(add_ptr, 2, 2, pointerarith_tfunc, 1)
+add_tfunc(sub_ptr, 2, 2, pointerarith_tfunc, 1)
 add_tfunc(pointerref, 3, 3, pointerref_tfunc, 4)
 add_tfunc(pointerset, 4, 4, pointerset_tfunc, 5)
 add_tfunc(atomic_fence, 1, 1, atomic_fence_tfunc, 4)

base/pointer.jl

@@ -313,8 +313,8 @@ isless(x::Ptr{T}, y::Ptr{T}) where {T} = x < y
 <(x::Ptr,  y::Ptr) = UInt(x) < UInt(y)
 -(x::Ptr,  y::Ptr) = UInt(x) - UInt(y)
 
-+(x::Ptr, y::Integer) = oftype(x, add_ptr(UInt(x), (y % UInt) % UInt))
--(x::Ptr, y::Integer) = oftype(x, sub_ptr(UInt(x), (y % UInt) % UInt))
++(x::Ptr, y::Integer) = add_ptr(x, (y % UInt) % UInt)
+-(x::Ptr, y::Integer) = sub_ptr(x, (y % UInt) % UInt)
 +(x::Integer, y::Ptr) = y + x
 
 unsigned(x::Ptr) = UInt(x)

base/task.jl

@@ -154,8 +154,7 @@ const _state_index = findfirst(==(:_state), fieldnames(Task))
 @eval function load_state_acquire(t)
     # TODO: Replace this by proper atomic operations when available
     @GC.preserve t llvmcall($("""
-        %ptr = inttoptr i$(Sys.WORD_SIZE) %0 to i8*
-        %rv = load atomic i8, i8* %ptr acquire, align 8
+        %rv = load atomic i8, i8* %0 acquire, align 8
         ret i8 %rv
         """), UInt8, Tuple{Ptr{UInt8}},
         Ptr{UInt8}(pointer_from_objref(t) + fieldoffset(Task, _state_index)))