feat(miner): scale displayed hashrate units dynamically

Add sha256_transform_armv8_2way_pass2 which reads the pass1 output state
words directly into MSG0/MSG1 without byte serialization. Previously:
  sha256_state_to_digest() → native uint32 → BE bytes (8x write_u32_be)
  sha256_transform load   → BE bytes → vrev32q_u8 → native uint32 (4x)
These two conversions cancel out. The new path skips both, saving ~52
shift/store/load/vrev ops per 4-nonce group. Also eliminates the two
128-byte block2 stack buffers from sha256d80_hash_4way_armv8_2way.

.gitignore

@@ -1,8 +1,31 @@
+# Python/local environment (prototype legacy)
 __pycache__/
+**/__pycache__/
 venv/
 .venv/
+.pytest_cache/
+*.pyc
+*.pyo
+
+# Local-only config
+miner.conf
 config.py
+
+# C/C++ build artifacts
 *.o
+*.a
+*.so
+
+# Binaries produced by Makefile
 miner
 launcher
-miner.conf
+bench_hash
+test_sha256_backend
+test_miner_regression
+
+# Profiling / PGO artifacts
+.pgo/
+*.gcda
+*.gcno
+*.profraw
+*.profdata

Makefile

@@ -1,10 +1,22 @@
 CC := gcc
 CFLAGS := -O3 -march=native -mtune=native -flto -fomit-frame-pointer -DNDEBUG -Wall -Wextra -std=c11 -D_GNU_SOURCE -D_POSIX_C_SOURCE=200809L
 LDFLAGS := -pthread -lcrypto -lm -flto
+PGO_DIR := .pgo
 
-COMMON_OBJS := config.o utils.o json.o rpc.o types.o block_builder.o miner.o mining_loop.o
+SHA256_BACKEND_OBJS := sha256/sha256_backend.o
+HAS_ARM_CRYPTO := $(shell echo | $(CC) $(CFLAGS) -dM -E - 2>/dev/null | grep -c __ARM_FEATURE_CRYPTO)
+ifeq ($(shell uname -m),aarch64)
+ifneq ($(HAS_ARM_CRYPTO),0)
+SHA256_BACKEND_OBJS += sha256/sha256d80_4way_aarch64.o
+endif
+endif
 
-.PHONY: all clean
+COMMON_OBJS := config.o utils.o json.o rpc.o types.o block_builder.o miner.o mining_loop.o $(SHA256_BACKEND_OBJS)
+BENCH_BIN := bench_hash
+TEST_SHA_BIN := test_sha256_backend
+TEST_MINER_BIN := test_miner_regression
+
+.PHONY: all clean bench bench-5x test pgo-gen pgo-use pgo-clean
 
 all: miner launcher
 
@@ -18,4 +30,39 @@ launcher: launcher.o $(COMMON_OBJS)
 	$(CC) $(CFLAGS) -c $< -o $@
 
 clean:
-	rm -f *.o miner launcher
+	rm -f *.o sha256/*.o bench/*.o tests/*.o miner launcher $(BENCH_BIN) $(TEST_SHA_BIN) $(TEST_MINER_BIN)
+
+bench: $(BENCH_BIN)
+
+$(BENCH_BIN): bench/bench_hash.o $(SHA256_BACKEND_OBJS)
+	$(CC) $(CFLAGS) -o $@ $^ $(LDFLAGS)
+
+bench-5x: $(BENCH_BIN)
+	@i=1; while [ $$i -le 5 ]; do \
+		echo "Run $$i/5"; \
+		./$(BENCH_BIN) 60; \
+		i=$$((i+1)); \
+	done
+
+$(TEST_SHA_BIN): tests/test_sha256_backend.o $(SHA256_BACKEND_OBJS) utils.o
+	$(CC) $(CFLAGS) -o $@ $^ $(LDFLAGS)
+
+$(TEST_MINER_BIN): tests/test_miner_regression.o miner.o $(SHA256_BACKEND_OBJS) utils.o
+	$(CC) $(CFLAGS) -o $@ $^ $(LDFLAGS)
+
+test: $(TEST_SHA_BIN) $(TEST_MINER_BIN)
+	./$(TEST_SHA_BIN)
+	./$(TEST_MINER_BIN)
+
+pgo-gen: CFLAGS += -fprofile-generate=$(PGO_DIR)
+pgo-gen: LDFLAGS += -fprofile-generate=$(PGO_DIR)
+pgo-gen: clean all
+	@echo "Profilazione build pronta."
+	@echo "Esegui il tuo workload manualmente, poi lancia: make pgo-use"
+
+pgo-use: CFLAGS += -fprofile-use=$(PGO_DIR) -fprofile-correction
+pgo-use: LDFLAGS += -fprofile-use=$(PGO_DIR)
+pgo-use: clean all
+
+pgo-clean:
+	rm -rf $(PGO_DIR)

bench/bench_hash.c

@@ -0,0 +1,59 @@
+#include <stdint.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+#include <time.h>
+
+#include "../sha256/sha256_backend.h"
+
+static double now_seconds(void) {
+    struct timespec ts;
+    clock_gettime(CLOCK_MONOTONIC, &ts);
+    return (double)ts.tv_sec + (double)ts.tv_nsec / 1e9;
+}
+
+int main(int argc, char **argv) {
+    uint8_t header_76[76];
+    sha256d80_midstate_t mid;
+    sha256_state_t states[4];
+    uint32_t target_words[8];
+    uint32_t nonce = 0;
+    uint64_t hashes = 0;
+    volatile uint32_t sink = 0;
+    double seconds = 10.0;
+    double t0;
+    int i;
+
+    if (argc > 1) {
+        seconds = strtod(argv[1], NULL);
+        if (seconds <= 0.0) {
+            seconds = 10.0;
+        }
+    }
+
+    for (i = 0; i < 76; i++) {
+        header_76[i] = (uint8_t)(i * 13 + 7);
+    }
+    for (i = 0; i < 8; i++) {
+        target_words[i] = 0xFFFFFFFFU;
+    }
+
+    sha256d80_midstate_init(&mid, header_76);
+
+    t0 = now_seconds();
+    while ((now_seconds() - t0) < seconds) {
+        uint32_t mask = sha256d80_scan_4way(&mid, nonce, target_words, states);
+
+        sink ^= states[0].h[0] ^ states[1].h[0] ^ states[2].h[0] ^ states[3].h[0] ^ mask;
+        nonce += 4U;
+        hashes += 4U;
+    }
+
+    {
+        double elapsed = now_seconds() - t0;
+        double khs = (double)hashes / elapsed / 1000.0;
+        printf("bench_hash: %.2f kH/s | hashes=%llu | sink=%08x\n", khs, (unsigned long long)hashes, sink);
+    }
+
+    return 0;
+}

launcher.c

@@ -53,6 +53,7 @@ static void dashboard_print(DashboardState *st, int n) {
     long long total_attempts = 0;
     time_t now_t = time(NULL);
     char ts[64];
+    char total_rate_buf[32];
 
     if (st->lines_printed > 0) {
         clear_lines(st->lines_printed);
@@ -64,13 +65,16 @@ static void dashboard_print(DashboardState *st, int n) {
     }
 
     strftime(ts, sizeof(ts), "%Y-%m-%d %H:%M:%S", localtime(&now_t));
+    format_hashrate_khs(total_rate, total_rate_buf, sizeof(total_rate_buf));
 
     printf("%s | MINING STATUS\n", ts);
     printf("========================================\n");
-    printf("Total: %.2f kH/s | Attempts: %lld\n", total_rate, total_attempts);
+    printf("Total: %s | Attempts: %lld\n", total_rate_buf, total_attempts);
     printf("----------------------------------------\n");
     for (i = 0; i < n; i++) {
-        printf("Worker %-2d: %.2f kH/s | Attempts: %lld\n", i, st->rates[i], st->attempts[i]);
+        char worker_rate_buf[32];
+        format_hashrate_khs(st->rates[i], worker_rate_buf, sizeof(worker_rate_buf));
+        printf("Worker %-2d: %s | Attempts: %lld\n", i, worker_rate_buf, st->attempts[i]);
     }
 
     st->lines_printed = 4 + n;
@@ -228,6 +232,8 @@ static int aggregate_loop(WorkerProc *workers, int n) {
             double elapsed = now_seconds() - st.start_t;
             long long total_attempts = 0;
             double avg_rate;
+            char winner_rate_buf[32];
+            char avg_rate_buf[32];
 
             if (st.lines_printed > 0) {
                 clear_lines(st.lines_printed);
@@ -238,6 +244,7 @@ static int aggregate_loop(WorkerProc *workers, int n) {
             }
 
             avg_rate = (elapsed > 0.0) ? (double)total_attempts / elapsed / 1000.0 : 0.0;
+            format_hashrate_khs(avg_rate, avg_rate_buf, sizeof(avg_rate_buf));
 
             printf("==============================================================================\n");
             printf("[OK] BLOCK FOUND AND SUBMITTED\n");
@@ -246,9 +253,10 @@ static int aggregate_loop(WorkerProc *workers, int n) {
                 printf("  Worker: %d\n", st.winner_idx);
             }
             if (st.has_winner_rate) {
-                printf("  Worker hashrate: %.2f kH/s\n", st.winner_rate);
+                format_hashrate_khs(st.winner_rate, winner_rate_buf, sizeof(winner_rate_buf));
+                printf("  Worker hashrate: %s\n", winner_rate_buf);
             }
-            printf("  Average total hashrate: %.2f kH/s\n", avg_rate);
+            printf("  Average total hashrate: %s\n", avg_rate_buf);
             printf("  Total attempts: %lld\n", total_attempts);
             printf("==============================================================================\n");
 

miner.c

@@ -1,99 +1,106 @@
 #include "miner.h"
 
-#include <openssl/sha.h>
 #include <stdio.h>
 #include <stdlib.h>
 #include <string.h>
 #include <time.h>
+#include <stddef.h>
 
+#include "sha256/sha256_backend.h"
 #include "utils.h"
 
-static const double RATE_INTERVAL_SEC = 5.0;
+static const double RATE_INTERVAL_SEC = 2.0;
 
-static int hash_meets_target(const uint8_t digest[32], const uint8_t target_be[32]) {
-    int i;
-
-    for (i = 0; i < 32; i++) {
-        uint8_t hb = digest[31 - i];
-        if (hb < target_be[i]) {
-            return 1;
-        }
-        if (hb > target_be[i]) {
-            return 0;
-        }
-    }
-
-    return 1;
+static inline uint32_t load_u32_be(const uint8_t *p) {
+    return ((uint32_t)p[0] << 24) |
+           ((uint32_t)p[1] << 16) |
+           ((uint32_t)p[2] << 8) |
+           (uint32_t)p[3];
 }
 
-static void write_u32_be(uint8_t *dst, uint32_t v) {
-    dst[0] = (uint8_t)((v >> 24) & 0xFF);
-    dst[1] = (uint8_t)((v >> 16) & 0xFF);
-    dst[2] = (uint8_t)((v >> 8) & 0xFF);
-    dst[3] = (uint8_t)(v & 0xFF);
-}
-
-static void sha256_ctx_to_digest(const SHA256_CTX *ctx, uint8_t out[32]) {
+static void target_words_from_be(const uint8_t target_be[32], uint32_t out[8]) {
     int i;
 
     for (i = 0; i < 8; i++) {
-        write_u32_be(out + i * 4, ctx->h[i]);
+        out[i] = load_u32_be(target_be + i * 4);
     }
 }
 
+static inline void write_u32_le(uint8_t *dst, uint32_t v) {
+    dst[0] = (uint8_t)(v & 0xFFU);
+    dst[1] = (uint8_t)((v >> 8) & 0xFFU);
+    dst[2] = (uint8_t)((v >> 16) & 0xFFU);
+    dst[3] = (uint8_t)((v >> 24) & 0xFFU);
+}
+
 static int compute_hash_batch(
     const uint8_t header_76[76],
     uint32_t start_nonce,
     uint32_t batch_size,
-    const uint8_t target_be[32],
+    const uint32_t target_words[8],
     uint32_t *found_nonce,
     uint8_t found_digest[32]
 ) {
-    SHA256_CTX init_ctx;
-    SHA256_CTX first_chunk_ctx;
-    uint8_t block1[64];
-    uint8_t block2[64];
-    uint8_t digest1[32];
-    uint8_t digest2[32];
-    uint32_t i;
+    sha256d80_midstate_t mid;
+    uint32_t i = 0;
+    sha256d80_midstate_init(&mid, header_76);
+    /* Backend is now initialized; use direct scan to skip per-call pthread_once */
+    sha256_backend_ensure_init();
 
-    SHA256_Init(&init_ctx);
-    first_chunk_ctx = init_ctx;
-    SHA256_Transform(&first_chunk_ctx, header_76);
+    while ((batch_size - i) >= 4U) {
+        uint32_t n0 = start_nonce + i;
+        sha256_state_t st1[4];
+        uint32_t mask = sha256d80_scan_4way_direct(&mid, n0, target_words, st1);
 
-    memset(block1, 0, sizeof(block1));
-    memcpy(block1, header_76 + 64, 12);
-    block1[16] = 0x80;
-    block1[62] = 0x02;
-    block1[63] = 0x80;
+        if (mask & 1U) {
+            *found_nonce = n0;
+            sha256_state_to_digest(&st1[0], found_digest);
+            return 1;
+        }
+        if (mask & 2U) {
+            *found_nonce = n0 + 1U;
+            sha256_state_to_digest(&st1[1], found_digest);
+            return 1;
+        }
+        if (mask & 4U) {
+            *found_nonce = n0 + 2U;
+            sha256_state_to_digest(&st1[2], found_digest);
+            return 1;
+        }
+        if (mask & 8U) {
+            *found_nonce = n0 + 3U;
+            sha256_state_to_digest(&st1[3], found_digest);
+            return 1;
+        }
 
-    memset(block2, 0, sizeof(block2));
-    block2[32] = 0x80;
-    block2[62] = 0x01;
-    block2[63] = 0x00;
+        i += 4U;
+    }
 
-    for (i = 0; i < batch_size; i++) {
-        SHA256_CTX ctx1;
-        SHA256_CTX ctx2;
-        uint32_t n = start_nonce + i;
+    if (i < batch_size) {
+        uint32_t n0 = start_nonce + i;
+        uint32_t remain = batch_size - i;
+        uint32_t valid_mask = (1U << remain) - 1U;
+        sha256_state_t st1[4];
+        uint32_t mask = sha256d80_scan_4way(&mid, n0, target_words, st1) & valid_mask;
 
-        block1[12] = (uint8_t)(n & 0xFF);
-        block1[13] = (uint8_t)((n >> 8) & 0xFF);
-        block1[14] = (uint8_t)((n >> 16) & 0xFF);
-        block1[15] = (uint8_t)((n >> 24) & 0xFF);
-
-        ctx1 = first_chunk_ctx;
-        SHA256_Transform(&ctx1, block1);
-        sha256_ctx_to_digest(&ctx1, digest1);
-
-        memcpy(block2, digest1, 32);
-        ctx2 = init_ctx;
-        SHA256_Transform(&ctx2, block2);
-        sha256_ctx_to_digest(&ctx2, digest2);
-
-        if (hash_meets_target(digest2, target_be)) {
-            *found_nonce = n;
-            memcpy(found_digest, digest2, 32);
+        if (mask & 1U) {
+            *found_nonce = n0;
+            sha256_state_to_digest(&st1[0], found_digest);
+            return 1;
+        }
+        if (mask & 2U) {
+            *found_nonce = n0 + 1U;
+            sha256_state_to_digest(&st1[1], found_digest);
+            return 1;
+        }
+        if (mask & 4U) {
+            *found_nonce = n0 + 2U;
+            sha256_state_to_digest(&st1[2], found_digest);
+            return 1;
+        }
+        if (mask & 8U) {
+            *found_nonce = n0 + 3U;
+            sha256_state_to_digest(&st1[3], found_digest);
             return 1;
         }
     }
@@ -101,13 +108,6 @@ static int compute_hash_batch(
     return 0;
 }
 
-static void write_u32_le(uint8_t *dst, uint32_t v) {
-    dst[0] = (uint8_t)(v & 0xFF);
-    dst[1] = (uint8_t)((v >> 8) & 0xFF);
-    dst[2] = (uint8_t)((v >> 16) & 0xFF);
-    dst[3] = (uint8_t)((v >> 24) & 0xFF);
-}
-
 int mine_block(
     uint8_t header_76[76],
     const char *target_hex,
@@ -120,12 +120,17 @@ int mine_block(
     MineResult *out
 ) {
     uint8_t target_be[32];
+    uint32_t target_words[8];
     uint32_t nonce;
     long long attempts = 0;
-    time_t t_start;
+    struct timespec ts_start, ts_now;
     double last_rate_t;
     long long last_rate_n = 0;
     double last_tsu;
+    uint32_t batch_count = 0;
+    /* Check timestamp update every 32 batches, rate every 32 batches (~1s) */
+#define TSU_CHECK_MASK  31U
+#define RATE_CHECK_MASK 31U
 
     memset(out, 0, sizeof(*out));
 
@@ -133,6 +138,7 @@ int mine_block(
         fprintf(stderr, "[miner] target hex non valido\n");
         return 0;
     }
+    target_words_from_be(target_be, target_words);
 
     if (strcmp(nonce_mode, "incremental") == 0) {
         nonce = 0;
@@ -140,30 +146,34 @@ int mine_block(
         nonce = (uint32_t)rand();
     }
 
-    t_start = time(NULL);
-    last_rate_t = (double)t_start;
-    last_tsu = (double)t_start;
+    clock_gettime(CLOCK_MONOTONIC_COARSE, &ts_start);
+    last_rate_t = (double)ts_start.tv_sec + (double)ts_start.tv_nsec * 1e-9;
+    last_tsu = last_rate_t;
 
     while (1) {
         uint32_t found_nonce = 0;
         uint8_t found_digest[32];
-        time_t now_t;
-        double now;
 
         if (atomic_load(stop_flag) != 0) {
             return 0;
         }
 
-        now_t = time(NULL);
-        now = (double)now_t;
+        batch_count++;
 
-        if (timestamp_update_interval > 0 && (now - last_tsu) >= (double)timestamp_update_interval) {
-            write_u32_le(header_76 + 68, (uint32_t)now_t);
-            last_tsu = now;
+        /* Timestamp update: check every 64 batches to reduce clock calls */
+        if (timestamp_update_interval > 0 && (batch_count & TSU_CHECK_MASK) == 0) {
+            clock_gettime(CLOCK_MONOTONIC_COARSE, &ts_now);
+            double now = (double)ts_now.tv_sec + (double)ts_now.tv_nsec * 1e-9;
+            if (now - last_tsu >= (double)timestamp_update_interval) {
+                write_u32_le(header_76 + 68, (uint32_t)ts_now.tv_sec);
+                last_tsu = now;
+            }
         }
 
-        if (compute_hash_batch(header_76, nonce, batch_size, target_be, &found_nonce, found_digest)) {
-            double total = now - (double)t_start;
+        if (compute_hash_batch(header_76, nonce, batch_size, target_words, &found_nonce, found_digest)) {
+            clock_gettime(CLOCK_MONOTONIC_COARSE, &ts_now);
+            double total = ((double)ts_now.tv_sec + (double)ts_now.tv_nsec * 1e-9)
+                         - ((double)ts_start.tv_sec + (double)ts_start.tv_nsec * 1e-9);
             if (total <= 0.0) {
                 total = 1e-6;
             }
@@ -182,18 +192,24 @@ int mine_block(
         attempts += batch_size;
         nonce += batch_size;
 
-        now = (double)time(NULL);
-        if ((now - last_rate_t) >= RATE_INTERVAL_SEC) {
-            double dt = now - last_rate_t;
-            if (dt <= 0.0) {
-                dt = 1e-6;
+        /* Rate reporting: check every 128 batches */
+        if ((batch_count & RATE_CHECK_MASK) == 0) {
+            clock_gettime(CLOCK_MONOTONIC_COARSE, &ts_now);
+            double now = (double)ts_now.tv_sec + (double)ts_now.tv_nsec * 1e-9;
+            if (now - last_rate_t >= RATE_INTERVAL_SEC) {
+                double dt = now - last_rate_t;
+                if (dt <= 0.0) {
+                    dt = 1e-6;
+                }
+                if (status_cb != NULL) {
+                    double hr = (double)(attempts - last_rate_n) / dt;
+                    status_cb(attempts, hr, status_ctx);
+                }
+                last_rate_t = now;
+                last_rate_n = attempts;
             }
-            if (status_cb != NULL) {
-                double hr = (double)(attempts - last_rate_n) / dt;
-                status_cb(attempts, hr, status_ctx);
-            }
-            last_rate_t = now;
-            last_rate_n = attempts;
         }
     }
+#undef TSU_CHECK_MASK
+#undef RATE_CHECK_MASK
 }

mining_loop.c

@@ -53,6 +53,7 @@ static void emit_eventf(int fd, const char *fmt, ...) {
 
 static void status_callback(long long attempts, double hashrate_hz, void *ctx_ptr) {
     StatusCtx *ctx = (StatusCtx *)ctx_ptr;
+    char rate_buf[32];
 
     if (ctx->event_fd >= 0) {
         emit_eventf(ctx->event_fd, "status %d %.6f %lld\n", ctx->worker_idx, hashrate_hz / 1000.0, attempts);
@@ -60,7 +61,8 @@ static void status_callback(long long attempts, double hashrate_hz, void *ctx_pt
     }
 
     if (ctx->single_mode) {
-        printf("\r[main] hashrate: %.2f kH/s | attempts: %lld", hashrate_hz / 1000.0, attempts);
+        format_hashrate_hz(hashrate_hz, rate_buf, sizeof(rate_buf));
+        printf("\r[main] hashrate: %s | attempts: %lld", rate_buf, attempts);
         fflush(stdout);
     }
 }

sha256/sha256_backend.h

@@ -0,0 +1,83 @@
+#ifndef SHA256_BACKEND_H
+#define SHA256_BACKEND_H
+
+#include <stdint.h>
+
+/*
+ * Compact SHA256 state: only the 8 chaining words (32 bytes).
+ * Avoids copying the bloated OpenSSL SHA256_CTX (~112 bytes) in the hot loop.
+ */
+typedef struct {
+    uint32_t h[8];
+} sha256_state_t;
+
+typedef struct {
+    sha256_state_t init_state;
+    sha256_state_t first_chunk_state;
+    uint8_t block1_template[64];
+    uint8_t block2_template[64];
+} sha256d80_midstate_t;
+
+/* Set state to SHA256 initial values (IV). */
+void sha256_state_init(sha256_state_t *state);
+
+/* Serialize the 8 state words into a 32-byte big-endian digest. */
+void sha256_state_to_digest(const sha256_state_t *state, uint8_t out[32]);
+
+/* Single SHA256 block compression (64-byte block). */
+void sha256_transform_fast(sha256_state_t *state, const uint8_t block[64]);
+
+/*
+ * 2-way interleaved SHA256 block compression.
+ * Processes two independent (state, block) pairs so the CPU can overlap both
+ * instruction chains. On non-ARM builds falls back to two sequential calls.
+ */
+void sha256_transform_fast_2way(
+    sha256_state_t *stA, const uint8_t blkA[64],
+    sha256_state_t *stB, const uint8_t blkB[64]
+);
+
+/* Prepare SHA256d(80-byte header) midstate and constant blocks from header[0..75]. */
+void sha256d80_midstate_init(sha256d80_midstate_t *mid, const uint8_t header_76[76]);
+
+/*
+ * Hash 4 consecutive nonces with SHA256d(header80).
+ * start_nonce lane order: [n, n+1, n+2, n+3].
+ */
+void sha256d80_hash_4way(
+    const sha256d80_midstate_t *mid,
+    uint32_t start_nonce,
+    sha256_state_t out_states[4]
+);
+
+/*
+ * Hash 4 consecutive nonces and return hit mask against target words.
+ * target_words are big-endian words target[0..7].
+ * bit i set => lane i meets target.
+ */
+uint32_t sha256d80_scan_4way(
+    const sha256d80_midstate_t *mid,
+    uint32_t start_nonce,
+    const uint32_t target_words[8],
+    sha256_state_t out_states[4]
+);
+
+/*
+ * Ensure the SHA256 backend is initialized. Call once before using
+ * sha256d80_scan_4way_direct() to avoid per-call pthread_once overhead.
+ */
+void sha256_backend_ensure_init(void);
+
+/*
+ * Like sha256d80_scan_4way() but skips the pthread_once check.
+ * Caller MUST have called sha256_backend_ensure_init() (or any other
+ * backend function) before calling this.
+ */
+uint32_t sha256d80_scan_4way_direct(
+    const sha256d80_midstate_t *mid,
+    uint32_t start_nonce,
+    const uint32_t target_words[8],
+    sha256_state_t out_states[4]
+);
+
+#endif

sha256/sha256d80_4way_aarch64.S

@@ -0,0 +1,9 @@
+.text
+.align 2
+
+.global sha256d80_4way_aarch64_kernel
+.type sha256d80_4way_aarch64_kernel, %function
+sha256d80_4way_aarch64_kernel:
+    b sha256d80_4way_aarch64_impl
+.size sha256d80_4way_aarch64_kernel, .-sha256d80_4way_aarch64_kernel
+

tests/test_miner_regression.c

@@ -0,0 +1,214 @@
+#include <stdatomic.h>
+#include <stdint.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+
+#include "../miner.h"
+#include "../sha256/sha256_backend.h"
+#include "../utils.h"
+
+static inline void write_u32_le(uint8_t *dst, uint32_t v) {
+    dst[0] = (uint8_t)(v & 0xFFU);
+    dst[1] = (uint8_t)((v >> 8) & 0xFFU);
+    dst[2] = (uint8_t)((v >> 16) & 0xFFU);
+    dst[3] = (uint8_t)((v >> 24) & 0xFFU);
+}
+
+static void sha256d_from_header76_nonce(const uint8_t header76[76], uint32_t nonce, uint8_t out[32]) {
+    uint8_t header80[80];
+    uint8_t block1[64];
+    uint8_t block2[64];
+    sha256_state_t init_state;
+    sha256_state_t st1;
+    sha256_state_t st2;
+
+    memcpy(header80, header76, 76);
+    write_u32_le(header80 + 76, nonce);
+
+    sha256_state_init(&init_state);
+    st1 = init_state;
+    sha256_transform_fast(&st1, header80);
+
+    memset(block1, 0, sizeof(block1));
+    memcpy(block1, header80 + 64, 16);
+    block1[16] = 0x80;
+    block1[62] = 0x02;
+    block1[63] = 0x80;
+    sha256_transform_fast(&st1, block1);
+
+    memset(block2, 0, sizeof(block2));
+    sha256_state_to_digest(&st1, block2);
+    block2[32] = 0x80;
+    block2[62] = 0x01;
+    block2[63] = 0x00;
+
+    st2 = init_state;
+    sha256_transform_fast(&st2, block2);
+    sha256_state_to_digest(&st2, out);
+}
+
+static void digest_rev(const uint8_t digest[32], uint8_t rev[32]) {
+    int i;
+    for (i = 0; i < 32; i++) {
+        rev[i] = digest[31 - i];
+    }
+}
+
+static int find_record_nonce(
+    const uint8_t header76[76],
+    int parity,
+    uint32_t min_nonce,
+    uint32_t max_nonce,
+    uint32_t *out_nonce,
+    uint8_t out_target_be[32]
+) {
+    uint8_t best[32];
+    uint32_t n;
+
+    memset(best, 0xFF, sizeof(best));
+
+    for (n = 0; n <= max_nonce; n++) {
+        uint8_t digest[32];
+        uint8_t rev[32];
+
+        sha256d_from_header76_nonce(header76, n, digest);
+        digest_rev(digest, rev);
+
+        if (memcmp(rev, best, 32) < 0) {
+            memcpy(best, rev, 32);
+            if (n >= min_nonce && ((int)(n & 1U) == parity)) {
+                *out_nonce = n;
+                memcpy(out_target_be, rev, 32);
+                return 1;
+            }
+        }
+    }
+
+    return 0;
+}
+
+static int run_case(uint32_t batch, int parity) {
+    uint8_t header76[76];
+    uint8_t target_be[32];
+    uint8_t expected_digest[32];
+    MineResult out;
+    atomic_int stop_flag;
+    uint32_t expected_nonce;
+    char *target_hex;
+    int rc;
+    int i;
+
+    for (i = 0; i < 76; i++) {
+        header76[i] = (uint8_t)((i * 29 + 11) & 0xFF);
+    }
+
+    if (!find_record_nonce(header76, parity, 4U, 150000U, &expected_nonce, target_be)) {
+        fprintf(stderr, "[test_miner_regression] no record nonce found for parity=%d\n", parity);
+        return 0;
+    }
+
+    target_hex = bytes_to_hex(target_be, 32);
+    if (target_hex == NULL) {
+        return 0;
+    }
+
+    atomic_init(&stop_flag, 0);
+    rc = mine_block(
+        header76,
+        target_hex,
+        "incremental",
+        batch,
+        0,
+        &stop_flag,
+        NULL,
+        NULL,
+        &out
+    );
+    free(target_hex);
+
+    if (rc == 0 || out.found == 0) {
+        fprintf(stderr, "[test_miner_regression] mine_block failed for batch=%u\n", batch);
+        return 0;
+    }
+
+    if (out.nonce != expected_nonce) {
+        fprintf(stderr, "[test_miner_regression] nonce mismatch batch=%u expected=%u got=%u\n",
+                batch, expected_nonce, out.nonce);
+        return 0;
+    }
+
+    sha256d_from_header76_nonce(header76, out.nonce, expected_digest);
+    if (memcmp(expected_digest, out.digest, 32) != 0) {
+        fprintf(stderr, "[test_miner_regression] digest mismatch for nonce=%u\n", out.nonce);
+        return 0;
+    }
+
+    return 1;
+}
+
+static int test_easy_target_finds_zero(void) {
+    uint8_t header76[76];
+    char *target_hex = NULL;
+    MineResult out;
+    atomic_int stop_flag;
+    int i;
+
+    for (i = 0; i < 76; i++) {
+        header76[i] = (uint8_t)((i * 7 + 3) & 0xFF);
+    }
+
+    target_hex = bytes_to_hex((const uint8_t[32]){
+        0xFF,0xFF,0xFF,0xFF,0xFF,0xFF,0xFF,0xFF,
+        0xFF,0xFF,0xFF,0xFF,0xFF,0xFF,0xFF,0xFF,
+        0xFF,0xFF,0xFF,0xFF,0xFF,0xFF,0xFF,0xFF,
+        0xFF,0xFF,0xFF,0xFF,0xFF,0xFF,0xFF,0xFF
+    }, 32);
+    if (target_hex == NULL) {
+        return 0;
+    }
+
+    atomic_init(&stop_flag, 0);
+    if (!mine_block(header76, target_hex, "incremental", 1, 0, &stop_flag, NULL, NULL, &out)) {
+        free(target_hex);
+        return 0;
+    }
+    free(target_hex);
+
+    return out.found == 1 && out.nonce == 0U;
+}
+
+static int test_wrap_arithmetic(void) {
+    uint32_t start = 0xFFFFFFFEU;
+    uint32_t n0 = start + 0U;
+    uint32_t n1 = start + 1U;
+    uint32_t n2 = start + 2U;
+    uint32_t n3 = start + 3U;
+
+    return n0 == 0xFFFFFFFEU &&
+           n1 == 0xFFFFFFFFU &&
+           n2 == 0x00000000U &&
+           n3 == 0x00000001U;
+}
+
+int main(void) {
+    if (!test_easy_target_finds_zero()) {
+        fprintf(stderr, "test_easy_target_finds_zero: FAIL\n");
+        return 1;
+    }
+    if (!run_case(5U, 1)) {
+        fprintf(stderr, "run_case(batch=5, parity=1): FAIL\n");
+        return 1;
+    }
+    if (!run_case(6U, 0)) {
+        fprintf(stderr, "run_case(batch=6, parity=0): FAIL\n");
+        return 1;
+    }
+    if (!test_wrap_arithmetic()) {
+        fprintf(stderr, "test_wrap_arithmetic: FAIL\n");
+        return 1;
+    }
+
+    printf("test_miner_regression: OK\n");
+    return 0;
+}

tests/test_sha256_backend.c

@@ -0,0 +1,266 @@
+#include <openssl/sha.h>
+#include <stdint.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+
+#include "../sha256/sha256_backend.h"
+#include "../utils.h"
+
+static uint32_t xorshift32(uint32_t *s) {
+    uint32_t x = *s;
+    x ^= x << 13;
+    x ^= x >> 17;
+    x ^= x << 5;
+    *s = x;
+    return x;
+}
+
+static inline void write_u32_le(uint8_t *dst, uint32_t v) {
+    dst[0] = (uint8_t)(v & 0xFFU);
+    dst[1] = (uint8_t)((v >> 8) & 0xFFU);
+    dst[2] = (uint8_t)((v >> 16) & 0xFFU);
+    dst[3] = (uint8_t)((v >> 24) & 0xFFU);
+}
+
+static void fill_random(uint8_t *dst, size_t n, uint32_t *seed) {
+    size_t i;
+    for (i = 0; i < n; i++) {
+        dst[i] = (uint8_t)(xorshift32(seed) & 0xFFU);
+    }
+}
+
+static void backend_double_sha256_80(const uint8_t header80[80], uint8_t out[32]) {
+    sha256_state_t init_state;
+    sha256_state_t st1;
+    sha256_state_t st2;
+    uint8_t block1[64];
+    uint8_t block2[64];
+
+    sha256_state_init(&init_state);
+
+    st1 = init_state;
+    sha256_transform_fast(&st1, header80);
+
+    memset(block1, 0, sizeof(block1));
+    memcpy(block1, header80 + 64, 16);
+    block1[16] = 0x80;
+    block1[62] = 0x02;
+    block1[63] = 0x80;
+    sha256_transform_fast(&st1, block1);
+
+    memset(block2, 0, sizeof(block2));
+    sha256_state_to_digest(&st1, block2);
+    block2[32] = 0x80;
+    block2[62] = 0x01;
+    block2[63] = 0x00;
+
+    st2 = init_state;
+    sha256_transform_fast(&st2, block2);
+    sha256_state_to_digest(&st2, out);
+}
+
+static int test_transform_equivalence_random(void) {
+    uint32_t seed = 0x12345678U;
+    int i;
+
+    for (i = 0; i < 10000; i++) {
+        uint8_t block[64];
+        SHA256_CTX ref;
+        sha256_state_t test;
+
+        fill_random(block, sizeof(block), &seed);
+
+        SHA256_Init(&ref);
+        sha256_state_init(&test);
+        SHA256_Transform(&ref, block);
+        sha256_transform_fast(&test, block);
+
+        if (memcmp(ref.h, test.h, sizeof(ref.h)) != 0) {
+            fprintf(stderr, "[test_sha256_backend] transform mismatch at iter=%d\n", i);
+            return 0;
+        }
+    }
+
+    return 1;
+}
+
+static int test_transform_2way_equivalence(void) {
+    uint32_t seed = 0xCAFEBABEU;
+    int i;
+
+    for (i = 0; i < 5000; i++) {
+        uint8_t block_a[64];
+        uint8_t block_b[64];
+        SHA256_CTX ref_a;
+        SHA256_CTX ref_b;
+        sha256_state_t test_a;
+        sha256_state_t test_b;
+
+        fill_random(block_a, sizeof(block_a), &seed);
+        fill_random(block_b, sizeof(block_b), &seed);
+
+        SHA256_Init(&ref_a);
+        SHA256_Init(&ref_b);
+        sha256_state_init(&test_a);
+        sha256_state_init(&test_b);
+
+        SHA256_Transform(&ref_a, block_a);
+        SHA256_Transform(&ref_b, block_b);
+        sha256_transform_fast_2way(&test_a, block_a, &test_b, block_b);
+
+        if (memcmp(ref_a.h, test_a.h, sizeof(ref_a.h)) != 0 ||
+            memcmp(ref_b.h, test_b.h, sizeof(ref_b.h)) != 0) {
+            fprintf(stderr, "[test_sha256_backend] 2way mismatch at iter=%d\n", i);
+            return 0;
+        }
+    }
+
+    return 1;
+}
+
+static int test_double_sha256_80_equivalence(void) {
+    uint32_t seed = 0xA55AA55AU;
+    int i;
+
+    for (i = 0; i < 2048; i++) {
+        uint8_t header80[80];
+        uint8_t ref[32];
+        uint8_t test[32];
+
+        fill_random(header80, sizeof(header80), &seed);
+        double_sha256(header80, sizeof(header80), ref);
+        backend_double_sha256_80(header80, test);
+
+        if (memcmp(ref, test, sizeof(ref)) != 0) {
+            fprintf(stderr, "[test_sha256_backend] sha256d80 mismatch at iter=%d\n", i);
+            return 0;
+        }
+    }
+
+    return 1;
+}
+
+static int test_sha256d80_4way_100k_nonces(void) {
+    uint8_t header76[76];
+    sha256d80_midstate_t mid;
+    uint32_t base = 0x12340000U;
+    int i;
+
+    for (i = 0; i < 76; i++) {
+        header76[i] = (uint8_t)((i * 19 + 5) & 0xFF);
+    }
+
+    sha256d80_midstate_init(&mid, header76);
+
+    for (i = 0; i < 25000; i++) {
+        uint32_t start_nonce = base + (uint32_t)(i * 4);
+        sha256_state_t states[4];
+        int lane;
+
+        sha256d80_hash_4way(&mid, start_nonce, states);
+
+        for (lane = 0; lane < 4; lane++) {
+            uint8_t header80[80];
+            uint8_t ref[32];
+            uint8_t got[32];
+            uint32_t nonce = start_nonce + (uint32_t)lane;
+
+            memcpy(header80, header76, 76);
+            write_u32_le(header80 + 76, nonce);
+            double_sha256(header80, sizeof(header80), ref);
+            sha256_state_to_digest(&states[lane], got);
+
+            if (memcmp(ref, got, 32) != 0) {
+                fprintf(stderr, "[test_sha256_backend] 4way nonce mismatch nonce=%u lane=%d\n", nonce, lane);
+                return 0;
+            }
+        }
+    }
+
+    return 1;
+}
+
+static int test_sha256d80_scan_4way_direct(void) {
+    uint8_t header76[76];
+    sha256d80_midstate_t mid;
+    sha256_state_t st_ref[4];
+    sha256_state_t st_direct[4];
+    uint32_t all_max[8];
+    uint32_t mask_ref;
+    uint32_t mask_direct;
+    int i;
+
+    for (i = 0; i < 76; i++) {
+        header76[i] = (uint8_t)((i * 13 + 7) & 0xFF);
+    }
+    for (i = 0; i < 8; i++) {
+        all_max[i] = 0xFFFFFFFFU;
+    }
+
+    sha256d80_midstate_init(&mid, header76);
+    sha256_backend_ensure_init();
+
+    mask_ref    = sha256d80_scan_4way(&mid, 0x11223344U, all_max, st_ref);
+    mask_direct = sha256d80_scan_4way_direct(&mid, 0x11223344U, all_max, st_direct);
+
+    if (mask_ref != mask_direct) {
+        fprintf(stderr, "[test_sha256_backend] scan_4way_direct mask mismatch ref=%u direct=%u\n",
+                mask_ref, mask_direct);
+        return 0;
+    }
+    if (memcmp(st_ref, st_direct, sizeof(st_ref)) != 0) {
+        fprintf(stderr, "[test_sha256_backend] scan_4way_direct state mismatch\n");
+        return 0;
+    }
+    return 1;
+}
+
+static int test_sha256d80_scan_hitmask_basic(void) {
+    uint8_t header76[76];
+    sha256d80_midstate_t mid;
+    sha256_state_t states[4];
+    uint32_t all_max[8];
+    uint32_t all_zero[8];
+    uint32_t mask_max;
+    uint32_t mask_zero;
+    int i;
+
+    for (i = 0; i < 76; i++) {
+        header76[i] = (uint8_t)((i * 11 + 1) & 0xFF);
+    }
+    for (i = 0; i < 8; i++) {
+        all_max[i] = 0xFFFFFFFFU;
+        all_zero[i] = 0x00000000U;
+    }
+
+    sha256d80_midstate_init(&mid, header76);
+    mask_max = sha256d80_scan_4way(&mid, 0xABC00000U, all_max, states);
+    mask_zero = sha256d80_scan_4way(&mid, 0xABC00000U, all_zero, states);
+
+    return mask_max == 0xFU && mask_zero == 0U;
+}
+
+int main(void) {
+    if (!test_transform_equivalence_random()) {
+        return 1;
+    }
+    if (!test_transform_2way_equivalence()) {
+        return 1;
+    }
+    if (!test_double_sha256_80_equivalence()) {
+        return 1;
+    }
+    if (!test_sha256d80_4way_100k_nonces()) {
+        return 1;
+    }
+    if (!test_sha256d80_scan_4way_direct()) {
+        return 1;
+    }
+    if (!test_sha256d80_scan_hitmask_basic()) {
+        return 1;
+    }
+
+    printf("test_sha256_backend: OK\n");
+    return 0;
+}

utils.c

@@ -15,6 +15,48 @@ static char to_hex_digit(unsigned int v) {
     return (char)('a' + (v - 10U));
 }
 
+static void format_hashrate_core(double rate_hs, char *out, size_t out_len) {
+    static const char *units[] = {"H/s", "kH/s", "MH/s", "GH/s", "TH/s", "PH/s", "EH/s"};
+    size_t unit_idx = 0;
+    const size_t unit_count = sizeof(units) / sizeof(units[0]);
+    double value;
+
+    if (out == NULL || out_len == 0) {
+        return;
+    }
+
+    if (!(rate_hs >= 0.0)) {
+        rate_hs = 0.0;
+    }
+    value = rate_hs;
+
+    while (value >= 999.95 && unit_idx + 1 < unit_count) {
+        value /= 1000.0;
+        unit_idx++;
+    }
+
+    while (value > 0.0 && value < 1.0 && unit_idx > 0) {
+        value *= 1000.0;
+        unit_idx--;
+    }
+
+    if (value >= 100.0) {
+        snprintf(out, out_len, "%.0f %s", value, units[unit_idx]);
+    } else if (value >= 10.0) {
+        snprintf(out, out_len, "%.1f %s", value, units[unit_idx]);
+    } else {
+        snprintf(out, out_len, "%.2f %s", value, units[unit_idx]);
+    }
+}
+
+void format_hashrate_hz(double rate_hz, char *out, size_t out_len) {
+    format_hashrate_core(rate_hz, out, out_len);
+}
+
+void format_hashrate_khs(double rate_khs, char *out, size_t out_len) {
+    format_hashrate_core(rate_khs * 1000.0, out, out_len);
+}
+
 void sb_init(StrBuf *sb) {
     sb->data = NULL;
     sb->len = 0;

utils.h

@@ -33,5 +33,7 @@ char *decode_nbits_hex(uint32_t nbits);
 char *calculate_target_hex(const char *bits_hex, double difficulty_factor, const char *network);
 
 char *hex_add_width(const char *base_hex, int add_value);
+void format_hashrate_hz(double rate_hz, char *out, size_t out_len);
+void format_hashrate_khs(double rate_khs, char *out, size_t out_len);
 
 #endif