bch/abla: make 128-bit muldiv MSVC-portable (fixes c2pool-bch Windows build)

abla.hpp evaluated the ABLA control-function muldiv(x,y,z)=x*y/z in a native
unsigned __int128 intermediate. __int128 is a GCC/Clang extension MSVC does not
provide (error C4235 __int128 keyword not supported on this architecture), so
c2pool-bch failed to compile on Windows while Linux/macOS passed -- blocking the
v0.2.1 c2pool-bch-windows package.

Keep the native __int128 path BYTE-EXACT for GCC/Clang (the shipping Linux/macOS
builds are unchanged) under #if defined(__SIZEOF_INT128__). For any compiler
without __int128 (MSVC), fall back to boost::multiprecision::uint128_t -- a fixed
128-bit unsigned type, header-only, already a project dependency (conan + system
libboost). ABLA math stays exact.

Guard: new bch_muldiv_kat_test pins the two paths equal. Layer A is a set of
cross-platform known answers (several with >2^64 products, so a 64-bit multiply
would fail) that muldiv must satisfy on every compiler incl. MSVC. Layer B, on
platforms with __int128, asserts muldiv_native == muldiv_portable across the full
ABLA operand space plus a deterministic 200k-vector wide fuzz -- proving the
portable arithmetic is bit-identical to native on the trusted platform, which
certifies the MSVC path we cannot run in this CI. Verified locally: KAT PASS
(native==portable over 150260 in-domain fuzz vectors, release+debug); ABLA growth
soak unchanged (mainnet final=41305530).

Wired bch_muldiv_kat_test into src/impl/bch/test/CMakeLists.txt and both
build.yml COIN_BCH target lists. p2pool-merged-v36 surface: NONE (ABLA is a local
build-time block-size budget only; no PoW/share/coinbase/PPLNS math touched).

Author: frstrtr

.github/workflows/build.yml

@@ -366,6 +366,7 @@ jobs:
                     bch_embedded_getwork_test bch_embedded_seam_workview_test \
                     bch_embedded_block_broadcast_test bch_coin_node_seam_test \
                     bch_asert_kat_test \
+                    bch_muldiv_kat_test \
                     bch_g1_oracle_byte_parity_test bch_g1_share_serialization_parity_test bch_g0_canonical_pin_test bch_g2_ratchet_gate_kat_test \
             -j8
 
@@ -460,6 +461,7 @@ jobs:
                     bch_embedded_getwork_test bch_embedded_seam_workview_test \
                     bch_embedded_block_broadcast_test bch_coin_node_seam_test \
                     bch_asert_kat_test \
+                    bch_muldiv_kat_test \
                     bch_g1_oracle_byte_parity_test bch_g1_share_serialization_parity_test bch_g0_canonical_pin_test bch_g2_ratchet_gate_kat_test \
             -j8
 

src/impl/bch/coin/abla.hpp

@@ -27,6 +27,10 @@
 #include <cstdint>
 #include <limits>
 
+// Portable 128-bit intermediate for muldiv on compilers without __int128 (MSVC).
+// Header-only; boost is already a c2pool dependency (conan + system libboost).
+#include <boost/multiprecision/cpp_int.hpp>
+
 namespace bch {
 namespace coin {
 namespace abla {
@@ -39,16 +43,42 @@ inline constexpr uint64_t MAX_CONSENSUS_BLOCK_SIZE     = uint64_t(2000) * ONE_ME
 // 2^7 fixed precision for the "asymmetry factor" (zeta). BCHN abla.h B7.
 inline constexpr uint64_t B7 = 1u << 7u;
 
-// muldiv(x,y,z) = x*y/z in 128-bit intermediate (BCHN abla.cpp). The platform
-// is gcc/clang C++20 on linux -> use the native __int128 path BCHN selects.
-inline uint64_t muldiv(uint64_t x, uint64_t y, uint64_t z) {
+// muldiv(x,y,z) = x*y/z evaluated in a 128-bit intermediate (BCHN abla.cpp).
+// Consensus-critical: the product x*y can reach ~2^128, so the intermediate MUST
+// be a true 128-bit type; the ABLA control function guarantees the final
+// quotient fits uint64_t (asserted below).
+//
+// GCC/Clang provide the native unsigned __int128 that BCHN selects -- that path
+// is kept BYTE-EXACT for the shipping Linux/macOS builds. MSVC has no __int128
+// (error C4235), which broke the c2pool-bch Windows package; on any compiler
+// without __int128 we fall back to boost::multiprecision::uint128_t, a fixed
+// 128-bit unsigned type that yields BIT-IDENTICAL results. The two paths are
+// pinned equal on Linux by muldiv_kat_test (native == portable across the full
+// ABLA operand space) -- the KAT guard required for this consensus math.
+inline uint64_t muldiv_portable(uint64_t x, uint64_t y, uint64_t z) {
+    assert(z != 0);
+    using u128 = boost::multiprecision::uint128_t;
+    const u128 res = (u128(x) * u128(y)) / u128(z);
+    assert(res <= u128(std::numeric_limits<uint64_t>::max()));
+    return static_cast<uint64_t>(res);
+}
+
+#if defined(__SIZEOF_INT128__)
+// Native path (GCC/Clang) -- unchanged; this is what the merged Linux/macOS
+// packages have always shipped. Exposed by name so the KAT can diff it against
+// muldiv_portable on the trusted platform.
+inline uint64_t muldiv_native(uint64_t x, uint64_t y, uint64_t z) {
     assert(z != 0);
     const unsigned __int128 res =
         (static_cast<unsigned __int128>(x) * static_cast<unsigned __int128>(y))
         / static_cast<unsigned __int128>(z);
     assert(res <= static_cast<unsigned __int128>(std::numeric_limits<uint64_t>::max()));
     return static_cast<uint64_t>(res);
 }
+inline uint64_t muldiv(uint64_t x, uint64_t y, uint64_t z) { return muldiv_native(x, y, z); }
+#else
+inline uint64_t muldiv(uint64_t x, uint64_t y, uint64_t z) { return muldiv_portable(x, y, z); }
+#endif
 
 // Algorithm configuration -- part of a chain's consensus params (BCHN abla.h).
 struct Config {

src/impl/bch/test/CMakeLists.txt

@@ -20,6 +20,7 @@ if(BUILD_TESTING)
         abla_growth_soak_test     # M5: dynamic ABLA budget-GROWTH proof above 32 MB floor (pure)
         coinbase_kat_segwit_predicate_test # KAT: DGB<->BCH coinbase pairing -- gated-segwit-OFF predicate pin (pure)
         asert_kat_test            # M5: ASERT DAA known-answer vs BCHN @89a591f gold properties + mainnet anchor (pure)
+        muldiv_kat_test           # ABLA: 128-bit muldiv native(__int128)==portable(boost uint128) equivalence -- MSVC-portability guard (pure)
         g1_oracle_byte_parity_test # G1: oracle byte-parity vs p2pool-merged-v36 bitcoincash[_testnet].py (pure)
         g1_share_serialization_parity_test # G1: share-serialization byte-parity vs jtoomim oracle wire (pure)
         g0_canonical_pin_test      # G0: canonical-pin vs p2poolBCH @6603b79 baseline + v35->v36 transition rule (pure)

src/impl/bch/test/muldiv_kat_test.cpp

@@ -0,0 +1,115 @@
+// ---------------------------------------------------------------------------
+// bch::abla muldiv 128-bit KAT -- the guard for the MSVC-portability rework.
+//
+// abla.hpp evaluates x*y/z (BCHN control function) in a 128-bit intermediate.
+// GCC/Clang use native unsigned __int128; MSVC (no __int128, error C4235) falls
+// back to boost::multiprecision::uint128_t. Both must be BIT-IDENTICAL or the
+// Windows package would silently diverge from the shipping Linux/macOS ABLA.
+//
+// This test pins that in two layers:
+//   A. cross-platform known-answers: muldiv() must produce exact fixed results
+//      on EVERY compiler (this is all MSVC can run -- no native to diff).
+//   B. Linux-only equivalence sweep (guarded by __SIZEOF_INT128__): the portable
+//      path must equal the native path across the full ABLA operand space plus a
+//      deterministic wide-operand fuzz. Proving portable == native on the trusted
+//      platform is what certifies the MSVC path we cannot run here.
+//
+// Build-INERT wrt consensus surface: pure header math, no node/RPC/boost-graph.
+// p2pool-merged-v36 surface: NONE (ABLA is a LOCAL build-time byte budget only).
+// ---------------------------------------------------------------------------
+
+#include <cstdint>
+#include <iostream>
+#include <limits>
+
+#include "../coin/abla.hpp"
+
+namespace {
+
+int failures = 0;
+#define CHECK(cond) do { if (!(cond)) { \
+    std::cerr << "FAIL: " #cond " @ line " << __LINE__ << "\n"; ++failures; } } while (0)
+
+using bch::coin::abla::muldiv;
+using bch::coin::abla::muldiv_portable;
+using bch::coin::abla::B7;
+
+constexpr uint64_t U64MAX = std::numeric_limits<uint64_t>::max();
+
+// Deterministic 64-bit LCG (no <random>, no Date/rand) -- reproducible vectors.
+struct Lcg {
+    uint64_t s;
+    explicit Lcg(uint64_t seed) : s(seed) {}
+    uint64_t next() { s = s * 6364136223846793005ULL + 1442695040888963407ULL; return s; }
+};
+
+} // namespace
+
+int main() {
+    // ---- A. cross-platform known answers (run on ALL compilers incl. MSVC) ----
+    // hand-computed; several exercise a product > 2^64 so the true 128-bit
+    // intermediate is required (a 64-bit multiply would wrap and fail these).
+    CHECK(muldiv(192, 16000000ULL, 128) == 24000000ULL);            // ABLA amplify shape
+    CHECK(muldiv(6, 7, 2) == 21ULL);
+    CHECK(muldiv(0, 12345, 7) == 0ULL);
+    CHECK(muldiv(U64MAX, 1, 1) == U64MAX);
+    CHECK(muldiv(U64MAX, U64MAX, U64MAX) == U64MAX);                // product ~2^128
+    CHECK(muldiv(1000000000000000000ULL, 1000000000ULL,
+                 1000000000000ULL) == 1000000000000000ULL);         // 1e27 intermediate
+    CHECK(muldiv(B7, 32ULL * 1000000ULL, B7) == 32ULL * 1000000ULL);// identity via B7
+
+    // portable path must independently satisfy the same known answers
+    CHECK(muldiv_portable(192, 16000000ULL, 128) == 24000000ULL);
+    CHECK(muldiv_portable(U64MAX, U64MAX, U64MAX) == U64MAX);
+    CHECK(muldiv_portable(1000000000000000000ULL, 1000000000ULL,
+                          1000000000000ULL) == 1000000000000000ULL);
+
+#if defined(__SIZEOF_INT128__)
+    using bch::coin::abla::muldiv_native;
+
+    // ---- B1. exact ABLA-domain vectors: native == portable ----
+    // operand shapes taken straight from State::NextBlockState call sites.
+    const uint64_t sizes[] = {
+        0, 1, B7, 192, 37938, 1000000ULL, 16ULL * 1000000ULL,
+        32ULL * 1000000ULL, 2000ULL * 1000000ULL, (1ULL << 40), (1ULL << 62)
+    };
+    for (uint64_t x : sizes)
+        for (uint64_t y : sizes)
+            for (uint64_t z : sizes) {
+                if (z == 0) continue;
+                // only compare where the true quotient fits uint64_t (the ABLA
+                // invariant); otherwise both paths take low-64 of an out-of-range
+                // value and the comparison is not meaningful.
+                const unsigned __int128 q =
+                    (static_cast<unsigned __int128>(x) * y) / z;
+                if (q > static_cast<unsigned __int128>(U64MAX)) continue;
+                CHECK(muldiv_native(x, y, z) == muldiv_portable(x, y, z));
+            }
+
+    // ---- B2. wide deterministic fuzz: native == portable ----
+    Lcg rng(0x9E3779B97F4A7C15ULL);
+    int compared = 0;
+    for (int i = 0; i < 200000; ++i) {
+        const uint64_t x = rng.next();
+        const uint64_t y = rng.next();
+        uint64_t z = rng.next();
+        if (z == 0) z = 1;
+        const unsigned __int128 q =
+            (static_cast<unsigned __int128>(x) * y) / z;
+        if (q > static_cast<unsigned __int128>(U64MAX)) continue; // keep in-domain
+        ++compared;
+        if (muldiv_native(x, y, z) != muldiv_portable(x, y, z)) {
+            std::cerr << "FAIL fuzz: x=" << x << " y=" << y << " z=" << z << "\n";
+            ++failures;
+        }
+    }
+    CHECK(compared > 1000); // sanity: the domain filter did not reject everything
+    std::cerr << "muldiv KAT: native==portable over " << compared << " fuzz vectors\n";
+#else
+    std::cerr << "muldiv KAT: no __int128 (MSVC) -- known-answer layer only\n";
+#endif
+
+    if (failures) { std::cerr << failures << " CHECK(s) FAILED\n"; return 1; }
+    std::cerr << "muldiv_kat_test: ALL PASS\n";
+    return 0;
+}