watl.v0.llvm.lottes.c: Did FArena

This commit is contained in:
2025-11-04 02:07:38 -05:00
parent f437be32e2
commit 81328819c6
2 changed files with 372 additions and 25 deletions

View File

@@ -136,6 +136,8 @@ enum { false = 0, true = 1, true_overflow, };
#define r_(ptr) cast(typeof_ptr(ptr)*R_, ptr)
#define v_(ptr) cast(typeof_ptr(ptr)*V_, ptr)
#define tr_(type, ptr) cast(type*R_, ptr)
#define tv_(type, ptr) cast(type*V_, ptr)
#define kilo(n) (cast(U8, n) << 10)
#define mega(n) (cast(U8, n) << 20)
@@ -220,11 +222,13 @@ typedef def_farray(B1, 2);
typedef def_farray(B1, 4);
typedef def_farray(B1, 8);
I_ U8 mem_copy (U8 dest, U8 src, U8 len) __asm__("memcpy");
I_ U8 mem_copy_overlapping(U8 dest, U8 src, U8 len) __asm__("memmove");
I_ B4 mem_zero (U8 dest, U8 len) __asm__("memset");
I_ U8 mem_copy (U8 dest, U8 src, U8 len) { return (U8)(__builtin_memcpy ((void*)dest, (void const*)src, len)); }
I_ U8 mem_copy_overlapping(U8 dest, U8 src, U8 len) { return (U8)(__builtin_memmove((void*)dest, (void const*)src, len)); }
I_ U8 mem_fill (U8 dest, U8 value, U8 len) { return (U8)(__builtin_memset ((void*)dest, (int) value, len)); }
I_ B4 mem_zero (U8 dest, U8 len) { if (dest == 0) return false; mem_fill(dest, 0, len); return true; }
#define struct_copy(type, dest, src) mem_copy(dest, src, sizeof(type))
#define struct_zero(type, dest) mem_zero(dest, sizeof(type))
I_ U8 align_pow2(U8 x, U8 b) {
assert(b != 0);
@@ -426,6 +430,38 @@ I_ Slice_Mem mem__shrink(AllocatorInfo ainfo, Slice_Mem mem, U8 size, Opts_mem_s
#pragma endregion Allocator Interface
#pragma region FArena (Fixed-Sized Arena)
typedef def_struct(Opts_farena) {
U8 alignment;
};
typedef def_struct(FArena) {
U8 start;
U8 capacity;
U8 used;
};
I_ void farena_init__u (U8 arena, U8 mem_ptr, U8 mem_len);
void farena__push__u (U8 arena, U8 amount, U8 type_width, U8 alignment, U8 slice_mem);
I_ void farena_reset__u (U8 arena);
I_ void farena_rewind__u(U8 arena, U8 sp_type_sig, U8 sp_slot);
I_ void farena_save__u (U8 arena, U8 sp);
I_ FArena farena_make (Slice_Mem mem);
I_ void farena_init (FArena_R arena, Slice_Mem byte);
I_ Slice_Mem farena__push (FArena_R arena, U8 amount, U8 type_width, Opts_farena*R_ opts);
I_ void farena_reset (FArena_R arena);
I_ void farena_rewind(FArena_R arena, AllocatorSP save_point);
I_ AllocatorSP farena_save (FArena arena);
void farena_allocator_proc(U8 data, U8 requested_size, U8 alignment, U8 old_ptr, U8 old_len, U4 op, /*AllocatorProc_Out*/U8 out);
#define ainfo_farena(arena) (AllocatorInfo){ .proc = farena_allocator_proc, .data = u8_(& arena) }
#define farena_push_mem(arena, amount, ...) farena__push(arena, amount, 1, opt_args(Opts_farena, lit(stringify(B1)), __VA_ARGS__))
#define farena_push(arena, type, ...) \
cast(type*, farena__push(arena, size_of(type), 1, opt_args(Opts_farena, __VA_ARGS__))).ptr
#define farena_push_array(arena, type, amount, ...) \
(Slice ## type){ farena__push(arena, size_of(type), amount, opt_args(Opts_farena, __VA_ARGS__)).ptr, amount }
#pragma endregion FArena
#pragma endregion Header
@@ -434,61 +470,59 @@ I_ Slice_Mem mem__shrink(AllocatorInfo ainfo, Slice_Mem mem, U8 size, Opts_mem_s
#pragma region Allocator Interface
I_ void allocator_query__u(U8 ainfo_proc, U8 ainfo_data, U8 allocator_query_info) {
assert(ainfo_proc != nullptr);
assert(ainfo_proc != null);
cast(AllocatorProc*, ainfo_proc)(ainfo_data, 0, 0, 0, 0, AllocatorOp_Query, allocator_query_info);
}
I_ void mem_free__u(U8 proc, U8 data, U8 mem_ptr, U8 mem_len) {
assert(proc != nullptr);
assert(proc != null);
cast(AllocatorProc*, proc)(data, 0, 0, mem_ptr, mem_len, AllocatorOp_Free, 0);
}
I_ void mem_reset__u(U8 proc, U8 data) {
assert(proc != nullptr);
assert(proc != null);
cast(AllocatorProc*, proc)(data, 0, 0, 0, 0, AllocatorOp_Reset, 0);
}
I_ void mem_rewind__u(U8 proc, U8 data, U8 sp_type_sig, U8 sp_slot) {
assert(proc != nullptr);
assert(proc != null);
cast(AllocatorProc*, proc)(data, 0, 0, sp_type_sig, sp_slot, AllocatorOp_Rewind, 0);
}
I_ void mem_save_point__u(U8 proc, U8 data, U8 sp) {
assert(proc != nullptr);
assert(proc != null);
uvar(AllocatorProc_Out, out) = {0};
cast(AllocatorProc*, proc)(data, 0, 0, 0, 0, AllocatorOp_SavePoint, u8(out));
struct_copy(AllocatorSP, sp, & out[offset_of(AllocatorProc_Out, save_point)]);
cast(AllocatorProc*, proc)(data, 0, 0, 0, 0, AllocatorOp_SavePoint, u8_(out));
struct_copy(AllocatorSP, sp, (U8) out + offset_of(AllocatorProc_Out, save_point));
}
I_ void mem__alloc__u(U8 out_mem, U8 proc, U8 data, U8 size, U8 alignment, B4 no_zero) {
assert(proc != nullptr);
assert(proc != null);
uvar(AllocatorProc_Out, out) = {0};
cast(AllocatorProc*, proc)(data, size, alignment, 0, 0, no_zero ? AllocatorOp_Alloc_NoZero : AllocatorOp_Alloc, u8_(out));
struct_copy(Slice_Mem, out_mem, & out[offset_of(AllocatorProc_Out, allocation)]);
struct_copy(Slice_Mem, out_mem, (U8) out + offset_of(AllocatorProc_Out, allocation));
}
I_ void mem__grow__u(U8 out_mem, U8 proc, U8 data, U8 old_ptr, U8 old_len, U8 size, U8 alignment, B4 no_zero, B4 give_actual) {
assert(proc != nullptr);
assert(proc != null);
uvar(AllocatorProc_Out, out) = {0};
cast(AllocatorProc*, proc)(data, size, alignment, old_ptr, old_len, no_zero ? AllocatorOp_Grow_NoZero : AllocatorOp_Grow, u8_(out));
if (give_actual == false) { u8_r(out)[offset_of(AllocatorProc_Out, allocation) + offset_of(Slice_Mem, len)] = size; }
struct_copy(Slice_Mem, out_mem, & out[offset_of(AllocatorProc_Out, allocation)]);
if (give_actual == false) { u8_r(out + offset_of(AllocatorProc_Out, allocation) + offset_of(Slice_Mem, len))[0] = size; }
struct_copy(Slice_Mem, out_mem, (U8) out + offset_of(AllocatorProc_Out, allocation));
}
I_ void mem__shrink__u(U8 out_mem, U8 proc, U8 data, U8 old_ptr, U8 old_len, U8 size, U8 alignment) {
assert(proc != nullptr);
assert(proc != null);
uvar(AllocatorProc_Out, out) = {0};
cast(AllocatorProc*, proc)(data, size, alignment, old_ptr, old_len, AllocatorOp_Shrink, u8_(out));
struct_copy(Slice_Mem, out_mem, & out[offset_of(AllocatorProc_Out, allocation)]);
struct_copy(Slice_Mem, out_mem, (U8) out + offset_of(AllocatorProc_Out, allocation));
}
I_ void mem__resize__u(U8 out_mem, U8 proc, U8 data, U8 old_ptr, U8 old_len, U8 size, U8 alignment, B4 no_zero, B4 give_acutal) {
Slice_Mem result;
if (old_len == size) { result = (Slice_Mem){old_ptr, old_len}; }
if (old_len < size) { mem__grow__u (u8_(& result), proc, data, old_ptr, old_len, size, alignment, no_zero, give_acutal); }
else { mem__shrink__u(u8_(& result), proc, data, old_ptr, old_len, size, alignment); }
struct_copy(Slice_Mem, out_mem, & result);
if (old_len == size) { struct_copy(Slice_Mem, out_mem, (U8)& slice_mem(old_ptr, old_len)); }
if (old_len < size) { mem__grow__u (out_mem, proc, data, old_ptr, old_len, size, alignment, no_zero, give_acutal); }
else { mem__shrink__u(out_mem, proc, data, old_ptr, old_len, size, alignment); }
}
I_ AllocatorQueryInfo allocator_query(AllocatorInfo ainfo) { AllocatorQueryInfo out; allocator_query__u(ainfo.proc, ainfo.data, & out); return out; }
I_ AllocatorQueryInfo allocator_query(AllocatorInfo ainfo) { AllocatorQueryInfo out; allocator_query__u(u8_(ainfo.proc), ainfo.data, u8_(& out)); return out; }
I_ void mem_free (AllocatorInfo ainfo, Slice_Mem mem) { mem_free__u (u8_(ainfo.proc), ainfo.data, mem.ptr, mem.len); }
I_ void mem_reset (AllocatorInfo ainfo) { mem_reset__u (u8_(ainfo.proc), ainfo.data); }
I_ void mem_rewind(AllocatorInfo ainfo, AllocatorSP save_point) { mem_rewind__u(u8_(ainfo.proc), ainfo.data, u8_(save_point.type_sig), save_point.slot); }
I_ AllocatorSP mem_save_point(AllocatorInfo ainfo) { AllocatorSP sp; mem_save_point__u(u8_(ainfo.proc), ainfo.data, u8(& sp)); return sp; }
I_ AllocatorSP mem_save_point(AllocatorInfo ainfo) { AllocatorSP sp; mem_save_point__u(u8_(ainfo.proc), ainfo.data, u8_(& sp)); return sp; }
I_ Slice_Mem mem__alloc(AllocatorInfo ainfo, U8 size, Opts_mem_alloc_R opts) {
assert(opts != nullptr); Slice_Mem result;
@@ -503,7 +537,7 @@ I_ Slice_Mem mem__grow(AllocatorInfo ainfo, Slice_Mem mem, U8 size, Opts_mem_gro
}
I_ Slice_Mem mem__resize(AllocatorInfo ainfo, Slice_Mem mem, U8 size, Opts_mem_resize_R opts) {
assert(opts != nullptr);
Slice_Mem out; mem__resize__u(& out, ainfo.proc, ainfo.data, mem.ptr, mem.len, size, opts->alignment, opts->no_zero, opts->give_actual);
Slice_Mem out; mem__resize__u(u8_(& out), u8_(ainfo.proc), ainfo.data, mem.ptr, mem.len, size, opts->alignment, opts->no_zero, opts->give_actual);
return out;
}
I_ Slice_Mem mem__shrink(AllocatorInfo ainfo, Slice_Mem mem, U8 size, Opts_mem_shrink_R opts) {
@@ -514,7 +548,120 @@ I_ Slice_Mem mem__shrink(AllocatorInfo ainfo, Slice_Mem mem, U8 size, Opts_mem_s
#pragma endregion Allocator Interface
#pragma region FArena (Fixed-Sized Arena)
I_ void farena_init__u(U8 arena, U8 mem_ptr, U8 mem_len) {
assert(arena != null);
u8_r(arena + offset_of(FArena, start) )[0] = mem_ptr;
u8_r(arena + offset_of(FArena, capacity))[0] = mem_len;
u8_r(arena + offset_of(FArena, used) )[0] = 0;
}
inline void farena__push__u(U8 arena, U8 amount, U8 type_width, U8 alignment, U8 result) {
if (amount == 0) { struct_zero(Slice_Mem, result); }
U8 reg desired = type_width * amount;
U8 reg to_commit = align_pow2(desired, alignment ? alignment : MEMORY_ALIGNMENT_DEFAULT);
U8_R reg used = u8_r(arena + offset_of(FArena, used));
U8 reg unused = u8_r(arena + offset_of(FArena, capacity))[0] - used[0]; assert(to_commit <= unused);
U8 reg ptr = u8_r(arena + offset_of(FArena, start) )[0] + used[0];
used[0] += to_commit;
struct_copy(Slice_Mem, result, (U8)& slice_mem(ptr, desired));
}
inline void farena__grow__u(U8 result, U8 arena, U8 old_ptr, U8 old_len, U8 requested_size, U8 alignment, B4 should_zero) {
assert(result != null);
assert(arena != null);
U8_R reg used = u8_r(arena + offset_of(FArena, used));
/*Check if the allocation is at the end of the arena*/{
U8 reg alloc_end = old_ptr + old_len;
U8 reg arena_end = u8_r(arena + offset_of(FArena, start))[0] + used[0];
if (alloc_end != arena_end) {
// Not at the end, can't grow in place
struct_zero(Slice_Mem, result);
return;
}
}
// Calculate growth
U8 reg grow_amount = requested_size - old_len;
U8 reg aligned_grow = align_pow2(grow_amount, alignment ? alignment : MEMORY_ALIGNMENT_DEFAULT);
U8 reg unused = u8_r(arena + offset_of(FArena, capacity))[0] - used[0];
if (aligned_grow > unused) {
// Not enough space
struct_zero(Slice_Mem, result);
return;
}
used[0] += aligned_grow;
struct_copy(Slice_Mem, result, (U8)& slice_mem(old_ptr, aligned_grow + requested_size));
memory_zero(old_ptr + old_len, grow_amount * cast(U8, should_zero));
}
inline void farena__shrink__u(U8 result, U8 arena, U8 old_ptr, U8 old_len, U8 requested_size, U8 alignment) {
assert(result != null);
assert(arena != null);
U8_R reg used = u8_r(arena + offset_of(FArena, used));
/*Check if the allocation is at the end of the arena*/ {
U8 reg alloc_end = old_ptr + old_len;
U8 reg arena_end = u8_r(arena + offset_of(FArena, start))[0] + used[0];
if (alloc_end != arena_end) {
// Not at the end, can't shrink but return adjusted size
struct_copy(Slice_Mem, result, (U8)& slice_mem(old_ptr, requested_size));
return;
}
}
U8 reg aligned_original = align_pow2(old_len, MEMORY_ALIGNMENT_DEFAULT);
U8 reg aligned_new = align_pow2(requested_size, alignment ? alignment : MEMORY_ALIGNMENT_DEFAULT);
used[0] -= (aligned_original - aligned_new);
struct_copy(Slice_Mem, result, (U8)& slice_mem(old_ptr, requested_size));
}
I_ void farena_reset__u(U8 arena) { u8_r(arena + offset_of(FArena, used))[0] = 0; }
I_ void farena_rewind__u(U8 arena, U8 sp_type_sig, U8 sp_slot) {
assert(sp_type_sig == (U8)& farena_allocator_proc);
U8 reg start = u8_r(arena + offset_of(FArena, start))[0];
U8_R reg used = u8_r(arena + offset_of(FArena, used));
U8 reg end = start + used[0]; assert_bounds(sp_slot, start, end);
used[0] -= sp_slot - start;
}
I_ void farena_save__u(U8 arena, U8 sp) {
u8_r(sp + offset_of(AllocatorSP, type_sig))[0] = (U8)& farena_allocator_proc;
u8_r(sp + offset_of(AllocatorSP, slot ))[0] = u8_r(arena + offset_of(FArena, used))[0];
}
void farena_allocator_proc(U8 arena, U8 requested_size, U8 alignment, U8 old_ptr, U8 old_len, U4 op, /*AllocatorProc_Out*/U8 out)
{
assert(out != null);
assert(arena != null);
U8 reg allocation = arena + offset_of(AllocatorProc_Out, allocation);
switch (op)
{
case AllocatorOp_Alloc:
case AllocatorOp_Alloc_NoZero:
farena__push__u(arena, requested_size, 1, alignment, allocation);
memory_zero(u8_r(allocation + offset_of(Slice_Mem, ptr))[0], u8_r(allocation + offset_of(Slice_Mem, len))[0] * op);
break;
case AllocatorOp_Free: break;
case AllocatorOp_Reset: farena_reset__u(arena); break;
case AllocatorOp_Grow:
case AllocatorOp_Grow_NoZero:
farena__grow__u(allocation, arena, old_ptr, old_len, requested_size, alignment, op - AllocatorOp_Grow_NoZero);
break;
case AllocatorOp_Shrink:
farena__shrink__u(allocation, arena, old_ptr, old_len, requested_size, alignment);
break;
case AllocatorOp_Rewind: farena_rewind__u(arena, old_ptr, old_len); break;
case AllocatorOp_SavePoint: farena_save__u(arena, allocation); break;
case AllocatorOp_Query:
u4_r(out + offset_of(AllocatorQueryInfo, features))[0] =
AllocatorQuery_Alloc
| AllocatorQuery_Reset
| AllocatorQuery_Resize
| AllocatorQuery_Rewind
;
U8 reg max_alloc = u8_r(arena + offset_of(FArena, capacity))[0] - u8_r(arena + offset_of(FArena, used))[0];
u8_r(out + offset_of(AllocatorQueryInfo, max_alloc))[0] = max_alloc;
u8_r(out + offset_of(AllocatorQueryInfo, min_alloc))[0] = 0;
u8_r(out + offset_of(AllocatorQueryInfo, left ))[0] = max_alloc;
farena_save__u(arena, out + offset_of(AllocatorQueryInfo, save_point));
break;
}
return;
}
#pragma endregion FArena
#pragma endregion Implementation