Sloppy mod and docs

doc/doc.adoc

@@ -578,6 +578,21 @@ pause = 1
 To make NESFab always pause on Microsoft Windows, first create a shortcut to the NESFab executable.
 Then, in the shortcut's properties, put `--pause` after the target path.
 
+=== `sloppy`
+
+This option improves compilation speed at the cost of program optimization. 
+It can be disabled on a per-function basis with the modifier <<mod_flags, `-sloppy`>>.
+
+*Command-line usage:*
+----
+nesfab --sloppy
+----
+
+*Configuration file usage:*
+----
+sloppy = 1
+----
+
 == Supported Mappers [[mappers]]
 
 NESFab supports a small set of https://www.nesdev.org/wiki/Mapper[mappers],
@@ -2600,6 +2615,7 @@ Unlike other programming languages, functions in NESFab cannot be nested or recu
 - <<mod_flags, `+graphviz`>>
 - <<mod_flags, `+info`>>
 - <<mod_flags, `+static`>>
+- <<mod_flags, `+sloppy`, `-sloppy`>>
 
 Example:
 ----
@@ -2700,6 +2716,7 @@ While the mode function is executing, NMIs will be handled using the supplied `n
 - <<mod_flags, `+graphviz`>>
 - <<mod_flags, `+info`>>
 - <<mod_flags, `+static`>>
+- <<mod_flags, `+sloppy`, `-sloppy`>>
 
 Example:
 ----
@@ -2773,6 +2790,7 @@ nmi identifier()
 - <<mod_flags, `+graphviz`>>
 - <<mod_flags, `+info`>>
 - <<mod_flags, `+static`>>
+- <<mod_flags, `+sloppy`, `-sloppy`>>
 
 *Why do NMI interrupt functions exist?*
 
@@ -2834,6 +2852,7 @@ irq identifier()
 - <<mod_flags, `+graphviz`>>
 - <<mod_flags, `+info`>>
 - <<mod_flags, `+static`>>
+- <<mod_flags, `+sloppy`, `-sloppy`>>
 
 === `asm` [[kw_asm]]
 
@@ -2879,7 +2898,6 @@ asm fn waste_time()
 - <<mod_data>>.
 - <<mod_vars>>.
 - <<mod_flags, `+zero_page`, `-zero_page`>>
-- <<mod_flags, `+graphviz`>>
 - <<mod_flags, `+info`>>
 
 The labels of an `asm` function are visible using the <<member_access, `.` operator>>.
@@ -3668,6 +3686,7 @@ The following flags exist:
              You must validate this yourself.
 - `palette_3`: Converts 4-byte palettes into 3-byte palettes.
 - `palette_25`: Converts 32-byte palettes into 25-byte palettes.
+- `+sloppy`, `-sloppy`: Enables / disables faster compilation speed, at the cost of performance.
 
 Example:
 ----

src/cg_isel.cpp

@@ -4811,7 +4811,7 @@ std::size_t select_instructions(log_t* log, fn_t& fn, ir_t& ir)
     static TLS rh::batman_map<cross_transition_t, result_t> rebuilt;
     static TLS std::vector<rh::apair<cross_cpu_t, isel_cost_t>> new_out_states;
 
-    bool const sloppy = compiler_options().sloppy;
+    bool const sloppy = fn.sloppy();
     unsigned const BASE_SEL_SIZE = sloppy ? 4 : 32;
     unsigned const BASE_MAP_SIZE = sloppy ? 8 : 128;
     auto const SELS_COST_BOUND = sloppy ? cost_fn(NOP_IMPLIED) : cost_fn(LDA_ABSOLUTE) * 2;

src/globals.cpp

@@ -720,6 +720,9 @@ fn_t::fn_t(global_t& global, type_t type, fn_def_t&& fn_def, std::unique_ptr<mod
     case FN_NMI:  m_pimpl.reset(new nmi_impl_t());  break;
     case FN_IRQ:  m_pimpl.reset(new irq_impl_t());  break;
     }
+
+    m_sloppy = compiler_options().sloppy || mod_test(this->mods(), MOD_sloppy);
+    m_sloppy &= !mod_test(this->mods(), MOD_sloppy, false);
 }
 
 fn_ht fn_t::mode_nmi() const
@@ -1189,7 +1192,7 @@ void fn_t::compile()
     } while(false)
 
         unsigned iter = 0;
-        unsigned const MAX_ITER = compiler_options().sloppy ? 10 : 100;
+        unsigned const MAX_ITER = sloppy() ? 10 : 100;
         bool changed;
 
         // Do this first, to reduce the size of the IR:

src/globals.hpp

@@ -401,6 +401,8 @@ friend class global_t;
     unsigned irq_index() const;
     xbitset_t<fn_ht> const& irq_used_in_modes() const; 
 
+    bool sloppy() const { return m_sloppy; }
+
     precheck_tracked_t const& precheck_tracked() const { assert(m_precheck_tracked); return *m_precheck_tracked; }
     auto const& precheck_group_vars() const { assert(m_precheck_group_vars); return m_precheck_group_vars; }
     auto const& precheck_parent_modes() const {assert(compiler_phase() > PHASE_PRECHECK); return m_precheck_parent_modes; }
@@ -518,6 +520,8 @@ friend class global_t;
     bool m_precheck_wait_nmi = false;
     bool m_precheck_fences = false;
 
+    bool m_sloppy = false;
+
     // Bitsets of all global vars read/written in fn (deep)
     // These get assigned by 'calc_reads_writes_purity'.
     // The thread synchronization is implicit in the order of compilation.

src/mods.inc

@@ -9,3 +9,4 @@ MOD(7,  static)
 MOD(8,  palette_3)
 MOD(9,  palette_25)
 MOD(10, sram)
+MOD(11, sloppy)

src/pass1.hpp

@@ -206,10 +206,10 @@ class pass1_t
         {
         default:      return 0;
         case FN_CT:   return 0;
-        case FN_FN:   return MOD_zero_page | MOD_align | MOD_inline | MOD_graphviz | MOD_static | MOD_info;
-        case FN_MODE: return MOD_zero_page | MOD_align | MOD_graphviz | MOD_static | MOD_info;
-        case FN_NMI:  return MOD_zero_page | MOD_align | MOD_graphviz | MOD_static | MOD_info;
-        case FN_IRQ:  return MOD_zero_page | MOD_align | MOD_graphviz | MOD_static | MOD_info;
+        case FN_FN:   return MOD_zero_page | MOD_align | MOD_inline | MOD_graphviz | MOD_static | MOD_info | MOD_sloppy;
+        case FN_MODE: return MOD_zero_page | MOD_align | MOD_graphviz | MOD_static | MOD_info | MOD_sloppy;
+        case FN_NMI:  return MOD_zero_page | MOD_align | MOD_graphviz | MOD_static | MOD_info | MOD_sloppy;
+        case FN_IRQ:  return MOD_zero_page | MOD_align | MOD_graphviz | MOD_static | MOD_info | MOD_sloppy;
         }
     }