[legacy/mid] Add scalar optimizations: fold, copy, GVN, GCM, SCCP, reassociation, DCE

1 files changed, 449 insertions, 0 deletions

diff --git a/rcc-legacy/src/mid/sccp.cc b/rcc-legacy/src/mid/sccp.cc
new file mode 100644
index 0000000..537773a
--- /dev/null
+++ b/rcc-legacy/src/mid/sccp.cc
@@ -0,0 +1,449 @@
+/* Source code for RCC-Legacy.
+   Copyright (C) 2026 Richard Qin
+
+This file is part of RCC.
+
+RCC is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free
+Software Foundation; either version 3, or (at your option) any later
+version.
+
+RCC is distributed in the hope that it will be useful, but WITHOUT ANY
+WARRANTY; without even the implied warranty of MERCHANTABILITY or
+FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
+for more details.
+
+You should have received a copy of the GNU General Public License
+along with RCC; see the file LICENSE.  If not see
+<http://www.gnu.org/licenses/>.  */
+
+/// @file sccp.cc
+/// @brief Sparse Conditional Constant Propagation (Wegman-Zadeck).
+///
+/// See docs/sccp.md for architectural notes.
+
+#include "rcc/all.h"
+
+#include <cassert>
+#include <cstdio>
+#include <set>
+#include <utility>
+#include <vector>
+
+namespace rcc {
+
+namespace {
+
+enum LatState : uchar { Top = 0, Const = 1, Bot = 2 };
+
+struct Lat {
+    LatState state;
+    Ref con;  // Only valid when state==Const, type==RCon
+};
+
+constexpr Lat kTop = {Top, {RTmp, 0}};
+constexpr Lat kBot = {Bot, {RTmp, 0}};
+
+bool con_eq(Fn* fn, Ref a, Ref b) {
+    if (req(a, b))
+        return true;
+    if (rtype(a) != RCon || rtype(b) != RCon)
+        return false;
+    if (a.val >= static_cast<uint>(fn->ncon) || b.val >= static_cast<uint>(fn->ncon))
+        return false;
+    Con* ca = &fn->con[a.val];
+    Con* cb = &fn->con[b.val];
+    if (ca->type != cb->type)
+        return false;
+    if (ca->type == Con::CAddr && !symeq(ca->sym, cb->sym))
+        return false;
+    return ca->bits.i == cb->bits.i && ca->flt == cb->flt;
+}
+
+bool lat_eq(Fn* fn, Lat a, Lat b) {
+    if (a.state != b.state)
+        return false;
+    if (a.state == Const)
+        return con_eq(fn, a.con, b.con);
+    return true;
+}
+
+// Check if unfoldable type.
+bool op_is_unfoldable(uint op) {
+    if (isstore(op))
+        return true;  // store does not define tmp, but safe
+    if (isload(op))
+        return true;
+    if (isalloc(op))
+        return true;
+    if (ispar(op))
+        return true;
+    if (isarg(op))
+        return true;
+    if (op == Ocall)
+        return true;
+    if (op == Oblit0 || op == Oblit1)
+        return true;
+    if (op == Ovastart || op == Ovaarg)
+        return true;
+    if (op == Odbgloc)
+        return true;
+    if (op == Osel0 || op == Osel1)
+        return true;
+    if (op == Onop)
+        return true;
+    return false;
+}
+
+struct Sccp {
+    Ctx* ctx;
+    Fn* fn;
+    std::vector<Lat> lat;             // ntmp
+    std::vector<char> blk_reachable;  // nblk
+    // edge_exec pairs.
+    std::set<std::pair<uint, uint>> edge_exec;
+    std::vector<Blk*> cfg_wl;
+    std::vector<uint> ssa_wl;  // tmp val queue
+
+    Lat arg_lat(Ref r) {
+        if (rtype(r) == RCon)
+            return {Const, r};
+        if (rtype(r) == RTmp) {
+            if (r.val >= lat.size())
+                return kBot;
+            return lat[r.val];
+        }
+        // RSlot / RType / RCall / RMem etc: do not participate in constant propagation
+        return kBot;
+    }
+
+    void set_lat(uint t, Lat nv) {
+        if (t >= lat.size() || t == 0)
+            return;
+        if (lat_eq(fn, lat[t], nv))
+            return;
+        // lattice monotonically descends: Top -> Const -> Bot
+        lat[t] = nv;
+        ssa_wl.push_back(t);
+    }
+
+    // Evaluate instruction using lattice.
+    Lat eval_ins(Ins* i) {
+        if (rtype(i->to) != RTmp)
+            return kBot;
+        if (op_is_unfoldable(i->op))
+            return kBot;
+
+        Lat a0 = arg_lat(i->arg[0]);
+        Lat a1;
+        if (req(i->arg[1], Ref::none()))
+            a1 = {Const, Ref::con_z()};  // phantom arg[1] for unary op
+        else
+            a1 = arg_lat(i->arg[1]);
+
+        // Ocopy special case: lattice equals arg[0]
+        if (i->op == Ocopy)
+            return a0;
+
+        if (a0.state == Top || a1.state == Top)
+            return kTop;
+        if (a0.state == Bot || a1.state == Bot)
+            return kBot;
+
+        // Both sides are Const: try to fold
+        Ins tmp = *i;
+        tmp.arg[0] = a0.con;
+        tmp.arg[1] = a1.con;
+        Ref r = foldref(ctx, fn, &tmp);
+        if (req(r, Ref::none()))
+            return kBot;
+        return {Const, r};
+    }
+
+    Lat eval_phi(Phi* p, Blk* in_blk) {
+        Lat acc = kTop;
+        for (uint k = 0; k < p->narg; k++) {
+            Blk* pred = p->blk[k];
+            if (!pred)
+                continue;
+            if (!edge_exec.count({pred->id, in_blk->id}))
+                continue;  // Incoming edge unreachable - skip
+            Lat a = arg_lat(p->arg[k]);
+            if (a.state == Bot)
+                return kBot;
+            if (a.state == Top)
+                continue;
+            // Const
+            if (acc.state == Top) {
+                acc = a;
+            } else if (!con_eq(fn, acc.con, a.con)) {
+                return kBot;
+            }
+        }
+        return acc;
+    }
+
+    // Activate CFG edge.
+    void activate_edge(Blk* pred, Blk* succ) {
+        if (!succ)
+            return;
+        auto key = std::make_pair(pred->id, succ->id);
+        if (!edge_exec.insert(key).second)
+            return;  // Already activated
+        // Enqueue succ if needed.
+        cfg_wl.push_back(succ);
+        if (!blk_reachable[succ->id])
+            blk_reachable[succ->id] = 1;
+    }
+
+    void eval_jmp(Blk* b) {
+        int t = b->jmp.type;
+        if (isret(t) || t == Jhlt)
+            return;
+        if (t == Jjmp) {
+            activate_edge(b, b->s1);
+            return;
+        }
+        if (INRANGE(t, Jjf, Jjf1)) {
+            // Conditional jumps.
+            activate_edge(b, b->s1);
+            activate_edge(b, b->s2);
+            return;
+        }
+        if (t == Jjnz) {
+            Lat c = arg_lat(b->jmp.arg);
+            if (c.state == Top)
+                return;
+            if (c.state == Bot) {
+                activate_edge(b, b->s1);
+                activate_edge(b, b->s2);
+                return;
+            }
+            // Const: check if non-zero
+            if (rtype(c.con) != RCon || c.con.val >= static_cast<uint>(fn->ncon)) {
+                activate_edge(b, b->s1);
+                activate_edge(b, b->s2);
+                return;
+            }
+            Con* cc = &fn->con[c.con.val];
+            if (cc->type != Con::CBits) {
+                // CAddr is treated as truthy (non-null address)
+                activate_edge(b, b->s1);
+                return;
+            }
+            if (cc->bits.i != 0)
+                activate_edge(b, b->s1);
+            else
+                activate_edge(b, b->s2);
+            return;
+        }
+        // Other unknown jmp types: activate both sides
+        activate_edge(b, b->s1);
+        activate_edge(b, b->s2);
+    }
+
+    void visit_block(Blk* b) {
+        // Re-eval phis.
+        for (Phi* p = b->phi; p; p = p->link) {
+            if (rtype(p->to) != RTmp)
+                continue;
+            Lat nv = eval_phi(p, b);
+            set_lat(p->to.val, nv);
+        }
+        for (uint k = 0; k < b->nins; k++) {
+            Ins* i = &b->ins[k];
+            if (rtype(i->to) != RTmp)
+                continue;
+            Lat nv = eval_ins(i);
+            set_lat(i->to.val, nv);
+        }
+        eval_jmp(b);
+    }
+
+    // Process tmp uses.
+    void visit_uses(uint t) {
+        if (t >= static_cast<uint>(fn->ntmp))
+            return;
+        Tmp* tt = &fn->tmp[t];
+        for (uint u = 0; u < tt->nuse; u++) {
+            Use& uu = tt->use[u];
+            if (uu.bid >= blk_reachable.size())
+                continue;
+            if (!blk_reachable[uu.bid])
+                continue;  // In unreachable block, ignore for now
+            Blk* ub = fn->rpo[uu.bid];
+            if (uu.type == Use::UPhi) {
+                Phi* p = uu.u.phi;
+                if (rtype(p->to) == RTmp) {
+                    Lat nv = eval_phi(p, ub);
+                    set_lat(p->to.val, nv);
+                }
+            } else if (uu.type == Use::UIns) {
+                Ins* i = uu.u.ins;
+                if (rtype(i->to) == RTmp) {
+                    Lat nv = eval_ins(i);
+                    set_lat(i->to.val, nv);
+                }
+            } else if (uu.type == Use::UJmp) {
+                eval_jmp(ub);
+            }
+        }
+    }
+
+    void solve() {
+        // Init state.
+        lat.assign(static_cast<size_t>(fn->ntmp), kTop);
+        blk_reachable.assign(static_cast<size_t>(fn->nblk), 0);
+        edge_exec.clear();
+        cfg_wl.clear();
+        ssa_wl.clear();
+
+        blk_reachable[fn->start->id] = 1;
+        cfg_wl.push_back(fn->start);
+        // Enqueue start.
+
+        while (!cfg_wl.empty() || !ssa_wl.empty()) {
+            // Process worklists.
+            while (!cfg_wl.empty()) {
+                Blk* b = cfg_wl.back();
+                cfg_wl.pop_back();
+                visit_block(b);
+            }
+            while (!ssa_wl.empty()) {
+                uint t = ssa_wl.back();
+                ssa_wl.pop_back();
+                visit_uses(t);
+            }
+        }
+    }
+
+    // Apply: Commit lattice to IR.
+    Ref maybe_const(Ref r) {
+        if (rtype(r) != RTmp)
+            return r;
+        if (r.val >= lat.size())
+            return r;
+        Lat& l = lat[r.val];
+        if (l.state == Const)
+            return l.con;
+        return r;
+    }
+
+    int apply() {
+        int n_ins = 0, n_phi = 0, n_jmp = 0, n_arg = 0;
+        for (Blk* b = fn->start; b; b = b->link) {
+            if (b->id >= blk_reachable.size() || !blk_reachable[b->id])
+                continue;
+
+            // Phi: Remove incoming edges from unreachable preds; replace arg operands
+            for (Phi* p = b->phi; p; p = p->link) {
+                uint w = 0;
+                for (uint k = 0; k < p->narg; k++) {
+                    Blk* pred = p->blk[k];
+                    if (!pred || !edge_exec.count({pred->id, b->id})) {
+                        // This incoming edge was not activated during SCCP -> unreachable -> delete
+                        continue;
+                    }
+                    Ref na = maybe_const(p->arg[k]);
+                    if (!req(na, p->arg[k]))
+                        n_arg++;
+                    if (w != k) {
+                        p->arg[w] = na;
+                        p->blk[w] = pred;
+                    } else {
+                        p->arg[w] = na;
+                    }
+                    w++;
+                }
+                p->narg = w;
+
+                // Keep constant phi.to for DCE.
+                if (rtype(p->to) == RTmp && p->to.val < lat.size() && lat[p->to.val].state == Const)
+                    n_phi++;
+            }
+
+            // Ins: Change Const def to Ocopy; replace arg operands
+            for (uint k = 0; k < b->nins; k++) {
+                Ins* i = &b->ins[k];
+                bool is_const_def = (rtype(i->to) == RTmp && i->to.val < lat.size() &&
+                                     lat[i->to.val].state == Const);
+                if (is_const_def && i->op != Ocopy && !op_is_unfoldable(i->op)) {
+                    Ref cr = lat[i->to.val].con;
+                    i->op = Ocopy;
+                    i->arg[0] = cr;
+                    i->arg[1] = Ref::none();
+                    n_ins++;
+                    continue;
+                }
+                // Otherwise just replace args
+                Ref na0 = maybe_const(i->arg[0]);
+                Ref na1 = maybe_const(i->arg[1]);
+                if (!req(na0, i->arg[0])) {
+                    i->arg[0] = na0;
+                    n_arg++;
+                }
+                if (!req(na1, i->arg[1])) {
+                    i->arg[1] = na1;
+                    n_arg++;
+                }
+            }
+
+            // jmp: Jjnz cond Const -> Jjmp; replace args
+            if (b->jmp.type == Jjnz) {
+                Lat c = arg_lat(b->jmp.arg);
+                if (c.state == Const && rtype(c.con) == RCon &&
+                    c.con.val < static_cast<uint>(fn->ncon)) {
+                    Con* cc = &fn->con[c.con.val];
+                    if (cc->type == Con::CBits) {
+                        // Choose succ
+                        Blk* taken = (cc->bits.i != 0) ? b->s1 : b->s2;
+                        b->jmp.type = Jjmp;
+                        b->jmp.arg = Ref::none();
+                        b->s1 = taken;
+                        b->s2 = nullptr;
+                        n_jmp++;
+                    } else if (cc->type == Con::CAddr) {
+                        // Address is always truthy
+                        b->jmp.type = Jjmp;
+                        b->jmp.arg = Ref::none();
+                        b->s1 = b->s1;
+                        b->s2 = nullptr;
+                        n_jmp++;
+                    }
+                } else {
+                    Ref na = maybe_const(b->jmp.arg);
+                    if (!req(na, b->jmp.arg)) {
+                        b->jmp.arg = na;
+                        n_arg++;
+                    }
+                }
+            } else if (!isret(b->jmp.type) && b->jmp.type != Jhlt && b->jmp.type != Jjmp) {
+                Ref na = maybe_const(b->jmp.arg);
+                if (!req(na, b->jmp.arg)) {
+                    b->jmp.arg = na;
+                    n_arg++;
+                }
+            }
+        }
+
+        if (debug['S']) {
+            FILE* out = ctx->dbg_buf ? ctx->dbg_buf : stderr;
+            fprintf(out, "> SCCP %s: const_ins=%d const_phi=%d jjnz_collapse=%d arg_subst=%d\n",
+                    fn->name, n_ins, n_phi, n_jmp, n_arg);
+        }
+        return n_ins + n_phi + n_jmp + n_arg;
+    }
+};
+
+}  // namespace
+
+void sccp(Ctx* ctx, Fn* fn) {
+    // Prerequisites: filluse and rpo ready.
+    Sccp s;
+    s.ctx = ctx;
+    s.fn = fn;
+    s.solve();
+    s.apply();
+}
+
+}  // namespace rcc