fix(pipeline): add S_WAIT_READ_RESP state to fix read-noise PRNG timing

- Emit rd_valid_o for exactly one combinational cycle before waiting for
  read response, ensuring the read-noise PRNG advances once per batch issue
- Fix query handshake: wait for query_ready before asserting query_valid
  to avoid valid&&ready handshake drops on clock edges
- Add dynamic timeout estimation in test utilities based on DUT parameters
- Update test-top Makefile to run all noise configurations by default
- Remove uv run prefix from cocotb-config Makefile invocation

.justfile

@@ -50,13 +50,6 @@ remote-dry cmd:
 
 # ── CAM verification ──────────────────────────────────────────────────────────
 
-# Run all 4 integrated CAM configurations (no-noise, write-only, read-only, combined)
-cam-test-all:
-    just remote "make -C hw/sim clean && make -C hw/sim test-top NUM_ROWS=64 LANES=8 HASH_BITS=512 WRITE_NOISE_EN=0 READ_NOISE_EN=0"
-    just remote "make -C hw/sim clean && make -C hw/sim test-top NUM_ROWS=64 LANES=8 HASH_BITS=512 WRITE_NOISE_EN=1 WRITE_NOISE_RATE_NUM=1 WRITE_NOISE_RATE_DEN=100 READ_NOISE_EN=0"
-    just remote "make -C hw/sim clean && make -C hw/sim test-top NUM_ROWS=64 LANES=8 HASH_BITS=512 WRITE_NOISE_EN=0 READ_NOISE_EN=1 READ_NOISE_RATE_NUM=1 READ_NOISE_RATE_DEN=100"
-    just remote "make -C hw/sim clean && make -C hw/sim test-top NUM_ROWS=64 LANES=8 HASH_BITS=512 WRITE_NOISE_EN=1 WRITE_NOISE_RATE_NUM=1 WRITE_NOISE_RATE_DEN=100 READ_NOISE_EN=1 READ_NOISE_RATE_NUM=1 READ_NOISE_RATE_DEN=100"
-
 # Run all CAM tests in one pass (model + top + all modules)
 cam-test-full:
     just remote "make -C hw/sim clean && make -C hw/sim test-all"

.stignore

@@ -15,3 +15,4 @@ outputs
 **/__pycache__
 **/sim_build
 **/__marimo__
+*.fst

hw/rtl/core/match_engine_pipeline.sv

@@ -31,9 +31,10 @@ module match_engine_pipeline (
     //==========================================================================
     // State encoding
     //==========================================================================
-    typedef enum logic [2:0] {
+    typedef enum logic [3:0] {
         S_IDLE,
         S_ISSUE_READ,
+        S_WAIT_READ_RESP,
         S_WAIT_SCORE,
         S_STAGE_CANDIDATES,
         S_PUSH_CANDIDATES,
@@ -373,6 +374,13 @@ module match_engine_pipeline (
                 end
 
                 S_ISSUE_READ: begin
+                    // Emit rd_valid_o for exactly one cycle (combinational), then
+                    // transition to S_WAIT_READ_RESP. This ensures the read-noise
+                    // PRNG in cam_read_noise only advances once per batch issue.
+                    state_q <= S_WAIT_READ_RESP;
+                end
+
+                S_WAIT_READ_RESP: begin
                     if (rd_valid_i) begin
                         // Capture staging data for popcount pipeline
                         rd_stage_valid_q      <= 1'b1;
@@ -380,8 +388,7 @@ module match_engine_pipeline (
                         rd_stage_hashes_q     <= rd_hashes_i;
                         rd_stage_lane_valid_q <= rd_lane_valid_i;
                         // Remember the lane mask for this batch (used in S_WAIT_SCORE
-                        // to determine batch_scores_done — survives until overwritten
+                        // to determine batch_scores_done).
-                        // by the next S_ISSUE_READ).
                         batch_lane_mask_q     <= rd_lane_valid_i;
                         issue_base_q          <= issue_base_q + `LANES;
                         issued_batches_q      <= issued_batches_q + 1;

hw/sim/Makefile

@@ -7,7 +7,7 @@ TOP_CONFIGS := no_noise write_noise read_noise
 help:
 	@echo "Available hw/sim targets:"
 	@echo "  make test-model"
-	@echo "  make test-top            # 只运行默认顶层配置 (no_noise)"
+	@echo "  make test-top             # 默认运行所有顶层噪声配置"
 	@echo "  make test-top-all         # 运行所有顶层噪声配置"
 	@echo "  make test-top-no_noise    # 无噪声集成测试"
 	@echo "  make test-top-write_noise # 写入噪声集成测试"
@@ -20,7 +20,7 @@ help:
 
 test-all: test-model test-top-all test-modules
 
-test-top: test-top-no_noise
+test-top: test-top-all
 
 test-top-all: $(TOP_CONFIGS:%=test-top-%)
 

hw/sim/mk/cocotb-common.mk

@@ -79,4 +79,4 @@ COMPILE_ARGS += -Wno-WIDTHEXPAND
 COMPILE_ARGS += -Wno-UNOPTFLAT
 endif
 
-include $(shell uv run cocotb-config --makefiles)/Makefile.sim
+include $(shell cocotb-config --makefiles)/Makefile.sim

hw/sim/tests/top/no_noise/test_no_noise.py

@@ -35,6 +35,7 @@ from model.ref_model import (
 from tests.top.utils import (
     collect_topk,
     dut_hash_bits,
+    dut_lanes,
     dut_num_rows,
     query_once,
     query_topk_once,
@@ -44,7 +45,6 @@ from tests.top.utils import (
     write_rows,
 )
 
-
 # ═══════════════════════════════════════════════════════════════════════════════
 # 编译冒烟测试 — 验证 cam_top 能正确 elaborate
 # ═══════════════════════════════════════════════════════════════════════════════
@@ -343,6 +343,8 @@ async def query_scan_blocks_writes_until_result_consumed(dut):
     dut.wr_valid.value = 0
 
     # Consume full serial stream so the DUT returns idle
-    beats = await collect_topk(dut, timeout_cycles=2000)
+    beats = await collect_topk(
+        dut, timeout_cycles=max(2000, (dut_num_rows(dut) // dut_lanes(dut)) * 24 + 100)
+    )
     assert len(beats) > 0
     assert beats[-1][3] == 1  # last asserted

hw/sim/tests/top/utils.py

@@ -170,14 +170,33 @@ async def collect_topk(dut, timeout_cycles: int = 2000):
     raise AssertionError("Top-K result stream did not finish")
 
 
-async def query_topk_once(dut, query, timeout_cycles=2000):
+# ── 默认超时估算 ──────────────────────────────────────────────────
+
+
+def dut_query_timeout_cycles(dut):
+    """基于 DUT 参数估算完整查询（扫描 + 串行结果输出）的超时周期数。
+
+    各通道串行输出一个 beat 需要约 24 个流水线周期；
+    总超时 = ceil(全部行数 / 通道数) * 24 + 固定裕量 2000 周期。
+    至少返回 2000 周期以防止极小配置下的不合理值。
+
+    Example: 4096 rows / 8 lanes → ceil(4096/8) * 24 + 2000 = 14288 cycles.
+    """
+    num_rows = dut_num_rows(dut)
+    lanes = dut_lanes(dut)
+    batches = (num_rows + lanes - 1) // lanes  # ceil division, no math import
+    return max(2000, batches * 24 + 2000)
+
+
+async def query_topk_once(dut, query, timeout_cycles=None):
     """发起一次查询并收集完整的串行 Top-K 结果流。
 
     完整流程：
       1. 等待 DUT 空闲
-      2. 通过 query_valid/query_ready 握手发送查询
+      2. 等待 query_ready 为高
-      3. 消费完整的结果流
+      3. 通过 query_valid/query_ready 握手发送查询
-      4. 读取 score_debug_flat（如果存在）
+      4. 消费完整的结果流
+      5. 读取 score_debug_flat（如果存在）
 
     返回：(beats, top1_index, top1_score, score_debug)
       - beats: [(rank, row, score, last), ...]
@@ -186,16 +205,21 @@ async def query_topk_once(dut, query, timeout_cycles=2000):
     await wait_idle(dut)
 
     dut.query_hash.value = int(query)
-    dut.query_valid.value = 1
 
-    # 等待查询握手完成
+    # 等待 query_ready 为高（DUT 已就绪），避免组合逻辑下降沿导致的
-    while True:
+    # valid&&ready 握手丢失问题
+    while not int(dut.query_ready.value):
         await RisingEdge(dut.clk)
-        if int(dut.query_ready.value):
-            break
 
+    # assert query_valid 覆盖一个上升沿完成握手
+    dut.query_valid.value = 1
+    await RisingEdge(dut.clk)
     dut.query_valid.value = 0
 
+    # 若调用者未指定超时，根据 DUT 参数动态估算
+    if timeout_cycles is None:
+        timeout_cycles = dut_query_timeout_cycles(dut)
+
     # 消费完整串行结果流
     beats = await collect_topk(dut, timeout_cycles=timeout_cycles)
 
@@ -213,12 +237,19 @@ async def query_topk_once(dut, query, timeout_cycles=2000):
     return beats, beats[0][1], beats[0][2], score_debug
 
 
-async def query_once(dut, query):
+async def query_once(dut, query, timeout_cycles=None):
     """发起查询，返回 (top1_index, top1_score, score_debug)。
 
     内部调用 query_topk_once 并消费完整结果流，仅保留 rank-0 数据。
+
+    Parameters
+    ----------
+    timeout_cycles : int or None
+        传递给 query_topk_once 的超时周期数。None 表示根据 DUT 参数动态估算。
     """
-    _, top1_index, top1_score, score_debug = await query_topk_once(dut, query)
+    _, top1_index, top1_score, score_debug = await query_topk_once(
+        dut, query, timeout_cycles=timeout_cycles,
+    )
     return top1_index, top1_score, score_debug