Master.h

/*
 * Copyright (c) 2017-2020, NVIDIA CORPORATION.  All rights reserved.
 * 
 * Licensed under the Apache License, Version 2.0 (the "License")
 * you may not use this file except in compliance with the License.
 * You may obtain a copy of the License at
 * 
 *     http://www.apache.org/licenses/LICENSE-2.0
 * 
 * Unless required by applicable law or agreed to in writing, software
 * distributed under the License is distributed on an "AS IS" BASIS,
 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 * See the License for the specific language governing permissions and
 * limitations under the License.
 */
#ifndef __AXI_T_MASTER__
#define __AXI_T_MASTER__

#include <systemc.h>
#include <ac_reset_signal_is.h>

#include <axi/axi4.h>
#include <nvhls_connections.h>
#include <hls_globals.h>

#include <queue>
#include <string>
#include <sstream>
#include <vector>
#include <map>
#include <math.h>
#include <boost/assert.hpp>

#include <boost/random/mersenne_twister.hpp>
#include <boost/random/uniform_int_distribution.hpp>
#include <algorithm>

struct masterCfg {
  enum {
    numWrites = 100,
    numReads = 100,
    readDelay = 0,
    seed = 0,
  };
  // enum only supports 32-bit types, so use static 64-bit types as needed (e.g., for >32bit addresses)
  static const uint64_t addrBoundLower = 0;
  static const uint64_t addrBoundUpper = 0xFFFFFFFF;
};

template <typename axiCfg, typename cfg>
class Master : public sc_module {
  BOOST_STATIC_ASSERT_MSG(axiCfg::useWriteResponses || cfg::numReads == 0 || cfg::readDelay != 0,
                "Must use a substantial read delay if reading without write responses");
 public:
  static const int kDebugLevel = 0;
  typedef axi::axi4<axiCfg> axi4_;

  typename axi4_::read::template master<> if_rd;
  typename axi4_::write::template master<> if_wr;

  sc_in<bool> reset_bar;
  sc_in<bool> clk;

  std::map<typename axi4_::Addr, typename axi4_::Data> localMem;
  std::map<typename axi4_::Addr, NVUINT8 > localMem_wstrb;
  std::vector<typename axi4_::Addr> validReadAddresses;
  std::vector<typename axi4_::Addr> validReadAddresses_q;
  std::vector< long long > validReadAddresses_ctr;

  static const int bytesPerBeat = axi4_::DATA_WIDTH >> 3;
  static const bool wResp = axiCfg::useWriteResponses;

  sc_out<bool> done;

  SC_CTOR(Master)
      : if_rd("if_rd"), if_wr("if_wr"), reset_bar("reset_bar"), clk("clk") {

    SC_THREAD(run);
    sensitive << clk.pos();
    async_reset_signal_is(reset_bar, false);
  }

 protected:
  void run() {
    static const int bytesPerWord = axi4_::DATA_WIDTH >> 3;
    static const int axiAddrBitsPerWord = nvhls::log2_ceil<bytesPerWord>::val;
    // Workaround for useBurst=0, which sets ALEN field to 0 width
    static const int ALEN_W = axiCfg::useBurst != 0 ? axi4_::ALEN_WIDTH : 1; 
    static const int WSTRB_W = axiCfg::useWriteStrobes != 0 ? axi4_::WSTRB_WIDTH : 1; 
    std::queue <typename axi4_::Addr> raddr_queue;
    std::queue < NVUINTW(ALEN_W) > rlen_queue;
    std::queue <typename axi4_::Addr> waddr_queue;
    std::queue < NVUINTW(ALEN_W) > wlen_queue;
    typename axi4_::Data check_data;
    NVUINT8 check_data_wstrb;
    NVUINT8 rcv_data_wstrb;
    typename axi4_::AddrPayload addr_pld;
    typename axi4_::ReadPayload data_pld;

    // Follow the same priority as nvhls_rand: environment, then preprocessor define, then config
    unsigned int seed = cfg::seed;
#ifdef RAND_SEED
    seed = (RAND_SEED);
#endif
    const char* env_rand_seed = std::getenv("RAND_SEED");
    if (env_rand_seed != NULL) seed = atoi(env_rand_seed);
    boost::random::mt19937 gen(seed);
    boost::random::uniform_int_distribution<uint64_t> random_addr(cfg::addrBoundLower, cfg::addrBoundUpper);
    boost::random::uniform_int_distribution<> random_wstrb(1, pow(2,WSTRB_W)-1);
    boost::random::uniform_int_distribution<> random_burstlen(0, axiCfg::maxBurstSize-1);
    boost::random::uniform_int_distribution<> uniform_rand;

    typename axi4_::Addr wr_addr = cfg::addrBoundLower;
    typename axi4_::Data wr_data = 0xf00dcafe12345678;
    NVUINTW(WSTRB_W) wstrb = ~0;
    NVUINTW(ALEN_W) wr_len = 0;
    typename axi4_::Addr rd_addr_next;
    typename axi4_::Addr rd_addr;
    NVUINTW(ALEN_W) rd_len;
    typename axi4_::AddrPayload wr_addr_pld;
    typename axi4_::WritePayload wr_data_pld;
    typename axi4_::WRespPayload wr_resp_pld;

    done = 0;
    unsigned int numWrites = 0;
    unsigned int numReads = 0;
    unsigned int numWritesOfBurst = 0;
    unsigned int numReadsOfBurst = 0;
    unsigned int numWriteResponses = 0;
    unsigned int numReadResponses = 0;
    bool writeInProgress = false;
    bool rd_conflict = false;
    bool wr_conflict = false;

    if_rd.reset();
    if_wr.reset();

    wait(20);
    while (1) {
      wait();

      // READ
      if (validReadAddresses.size() > 0 && numReads < cfg::numReads) {
        rd_addr_next = validReadAddresses[uniform_rand(gen) % validReadAddresses.size()];
        addr_pld.addr = rd_addr_next;
        if (axiCfg::useBurst) {
          if (uniform_rand(gen) % 2 == 0) { // 50% of reads are bursts (less once valid addresses are disallowed)
            rd_len = random_burstlen(gen);
          } else {
            rd_len = 0;
          }
          addr_pld.len = rd_len;
        }
        wr_conflict = false;
        for (unsigned int i=0; i<(rd_len+1); i++) {
          if (!localMem.count(rd_addr_next+bytesPerBeat*i)) {
            wr_conflict = true; // Not actually a conflict, but the read should be cancelled nonetheless
          }
        }
        std::ostringstream ms3;
        ms3 << "\nError @" << sc_time_stamp() << " from " << name()
            << ": Testharness attempted to read an address that it never wrote to"
            << ", read_addr=" << hex << rd_addr_next
            << endl;
        BOOST_ASSERT_MSG( localMem.count(rd_addr_next), ms3.str().c_str() );
        for (unsigned int j=0; j<waddr_queue.size(); j++) {
          if ((rd_addr_next+bytesPerBeat*rd_len) >= waddr_queue.front() && rd_addr_next <= waddr_queue.front())
                wr_conflict = true;
          waddr_queue.push(waddr_queue.front());
          waddr_queue.pop();
        }
        if (!wr_conflict) {
          rd_addr_next = addr_pld.addr;
          if (if_rd.ar.PushNB(addr_pld)) {
            CDCOUT(sc_time_stamp() << " " << name() << " Sent read request: ["
                          << addr_pld << "]"
                          << endl, kDebugLevel);
            numReads++;
            for (unsigned int i=0; i<(rd_len+1); i++) {
              raddr_queue.push(rd_addr_next);
              rd_addr_next += bytesPerBeat;
            }
            rlen_queue.push(rd_len);
          }
        }
      }
      if (if_rd.r.PopNB(data_pld)) {
        rd_addr = raddr_queue.front();
        if (axiCfg::useWriteStrobes) {
          bool checked_one = false;
          for (int i=0; i<axi4_::WSTRB_WIDTH; i++) {
            if (localMem_wstrb.count(rd_addr+i)) {
              rcv_data_wstrb = nvhls::get_slc<8>(data_pld.data,8*i);
              check_data_wstrb = localMem_wstrb[rd_addr+i];
              std::ostringstream msg;
              msg << "\nError @" << sc_time_stamp() << " from " << name()
                  << ": Incorrect read data response"
                  << ", base_addr=" << hex << rd_addr
                  << ", byte=" << i
                  << ", data=" << hex << rcv_data_wstrb
                  << ", expected=" << hex << check_data_wstrb
                  << std::endl;
              BOOST_ASSERT_MSG( check_data_wstrb == rcv_data_wstrb, msg.str().c_str() );
              checked_one = true;
            }
          } 
          std::ostringstream msg3;
          msg3 << "\nError @" << sc_time_stamp() << " from " << name()
              << ": Testbench did not check any bytes of a read response" << std::endl;
          BOOST_ASSERT_MSG( checked_one, msg3.str().c_str() );
        } else {
          check_data = localMem[rd_addr];
          std::ostringstream msg;
          msg << "\nError @" << sc_time_stamp() << " from " << name()
              << ": Incorrect read data response"
              << ", addr=" << hex << rd_addr
              << ", data=" << hex << data_pld.data.to_uint64()
              << ", expected=" << hex << check_data.to_uint64()
              << std::endl;
          BOOST_ASSERT_MSG( check_data == data_pld.data , msg.str().c_str() );
        }
        std::ostringstream ms2;
        ms2 << "\nError @" << sc_time_stamp() << " from " << name()
            << ": Read response protocol error"
            << ", rresp=" << data_pld.resp.to_uint64()
            << std::endl;
        BOOST_ASSERT_MSG( (data_pld.resp == axi4_::Enc::XRESP::OKAY) |
                          (data_pld.resp == axi4_::Enc::XRESP::EXOKAY), ms2.str().c_str() );
        CDCOUT(sc_time_stamp() << " " << name() << " Received correct read response: ["
                      << data_pld << "]"
                      << endl, kDebugLevel);
        raddr_queue.pop();
        if (numReadsOfBurst++ == rlen_queue.front()) {
          numReadsOfBurst = 0;
          numReadResponses++;
          rlen_queue.pop();
        }
      }

      // WRITE
      wr_addr_pld.addr = wr_addr;
      wr_data_pld.data = wr_data;
      wr_data_pld.wstrb = wstrb;
      wr_addr_pld.len = wr_len;
      if (!writeInProgress && numWrites < cfg::numWrites) {
        rd_conflict = false;
        for (unsigned int j=0; j<raddr_queue.size(); j++) {
          if ((wr_addr+bytesPerBeat*wr_len) >= raddr_queue.front() && wr_addr <= raddr_queue.front())
                rd_conflict = true;
          raddr_queue.push(raddr_queue.front());
          raddr_queue.pop();
        }
        if (!rd_conflict) {
          if (if_wr.aw.PushNB(wr_addr_pld)) {
            CDCOUT(sc_time_stamp() << " " << name() << " Sent write request: ["
                          << wr_addr_pld << "]"
                          << endl, kDebugLevel);
            for (unsigned int i=0; i<(wr_len+1); i++) {
              waddr_queue.push(wr_addr+bytesPerBeat*i);
              // If the address was already written to once, it needs to be removed from the list of valid addresses
              // because a read after this second write could return incorrect data
              validReadAddresses.erase(std::remove(validReadAddresses.begin(), validReadAddresses.end(), waddr_queue.front()), validReadAddresses.end());
              if (!wResp) {
                ptrdiff_t pos = find(validReadAddresses_q.begin(), validReadAddresses_q.end(), waddr_queue.front()) - validReadAddresses_q.begin();
                int size = validReadAddresses_q.size();
                // If the address is already in the queue, just reset the counter
                if (pos < size) {
                  validReadAddresses_ctr[pos] = cfg::readDelay + 20*i;
                } else {
                  validReadAddresses_q.push_back(waddr_queue.front());
                  validReadAddresses_ctr.push_back(cfg::readDelay + 20*i);
                }
                waddr_queue.pop();
              }
            }
            wlen_queue.push(wr_len);
            writeInProgress = true;
          }
        }
      }
      if (writeInProgress) {
        if (axiCfg::useBurst) {
          if (numWritesOfBurst == (wr_len)) {
            wr_data_pld.last = 1;
          } else {
            wr_data_pld.last = 0;
          }
        }
        if (if_wr.w.PushNB(wr_data_pld)) {
          CDCOUT(sc_time_stamp() << " " << name() << " Sent write data:"
                        << " addr=" << hex << wr_addr
                        << " data=[" << wr_data_pld << "]"
                        << " beat=" << dec << numWritesOfBurst
                        << endl, kDebugLevel);
          if (axiCfg::useWriteStrobes) {
            for (int i=0; i<axi4_::WSTRB_WIDTH; i++) {
              if (wr_data_pld.wstrb[i] == 1) {
                localMem_wstrb[wr_addr+i] = nvhls::get_slc<8>(wr_data_pld.data, 8*i);
              }
            }
          }
          localMem[wr_addr] = wr_data_pld.data; // Need to keep track of base addresses, even for wstrb case
          if (++numWritesOfBurst == (wr_len+1)) { // Whole burst is done
            wr_addr = (random_addr(gen) >> axiAddrBitsPerWord) << axiAddrBitsPerWord; // Keep all requests word-aligned
            if (axiCfg::useBurst) {
              if (uniform_rand(gen) % 5 == 0) { // 20% of writes are bursts
                wr_len = random_burstlen(gen);
                if (wr_addr + bytesPerBeat*wr_len > cfg::addrBoundUpper) wr_len = 0;
              } else {
                wr_len = 0;
              }
            }
            writeInProgress = false;
            numWrites++;
            numWritesOfBurst = 0;
          } else { // Only this beat is done
            wr_addr += bytesPerBeat;
          }
          wr_data = 0xf00dcafe12345678
                      ^ ((static_cast<typename axi4_::Data>(wr_addr)) << 16) // Typically touches bits 47:16
                      ^ uniform_rand(gen); // Touches the 32 LSBs
          wr_data.set_slc(axi4_::DATA_WIDTH-8,NVUINT8(uniform_rand(gen))); // Touches the 8 MSBs, in case of wide data words
          if (axiCfg::useWriteStrobes) {
            if (uniform_rand(gen) % 5 == 0) { // 20% of writes have nonuniform strobe
              wstrb = random_wstrb(gen);
            } else {
              wstrb = ~0;
            }
          }
        }
      }
      if (if_wr.b.PopNB(wr_resp_pld)) {
        std::ostringstream msg;
        msg << "\nError @" << sc_time_stamp() << " from " << name()
            << ":  Write response protocol error"
            << ", bresp=" << wr_resp_pld.resp.to_uint64()
            << ", addr=" << hex << waddr_queue.front()
            << std::endl;
        BOOST_ASSERT_MSG( (wr_resp_pld.resp == axi4_::Enc::XRESP::OKAY) |
                          (wr_resp_pld.resp == axi4_::Enc::XRESP::EXOKAY), msg.str().c_str() );
        CDCOUT(sc_time_stamp() << " " << name() << " Received write response"
                      << endl, kDebugLevel);
        numWriteResponses++;
        for (unsigned int i=0; i<wlen_queue.front()+1; i++) {
          validReadAddresses_q.push_back(waddr_queue.front());
          validReadAddresses_ctr.push_back(cfg::readDelay + 20*i);
          waddr_queue.pop();
        }
        wlen_queue.pop();
      }
      if (numWrites == cfg::numWrites &&
          numReads == cfg::numReads &&
          (numWriteResponses == numWrites || !wResp) &&
          numReadResponses == numReads)
            done = 1;
      //DCOUT(sc_time_stamp() << " from " << name() << " numWrites: " << dec << numWrites << "  numWriteResponses: " << numWriteResponses << "  numReads: " << numReads << "  numReadResponses: " << numReadResponses << endl);
      std::ostringstream msg;
      msg << "\nError @" << sc_time_stamp() << " from " << name()
          << ": Number of write responses exceeded total number of writes"
          << ", numWriteResponses=" << dec << numWriteResponses
          << std::endl;
      BOOST_ASSERT_MSG( numWriteResponses <= cfg::numWrites, msg.str().c_str() );
      std::ostringstream ms2;
      ms2 << "\nError @" << sc_time_stamp() << " from " << name()
          << ": Number of read responses exceeded total number of reads"
          << ", numReadResponses=" << dec << numReadResponses
          << std::endl;
      BOOST_ASSERT_MSG( numReadResponses <= cfg::numReads, msg.str().c_str() );

      // If there are no write responses we still want to test reads.
      // Finesse this by enforcing a fixed delay after a write until reads can be issued to that address.
      for (unsigned int i=0; i<validReadAddresses_q.size(); i++) {
        if (validReadAddresses_ctr[i] >= 0) {
          if (validReadAddresses_ctr[i] == 0) {
            validReadAddresses.push_back(validReadAddresses_q[i]);
          }
          validReadAddresses_ctr[i]--;
        }
      }
      //DCOUT("validReadAddresses: ");
      //for (typename std::vector< sc_uint<axi4_::ADDR_WIDTH> >::const_iterator q = validReadAddresses.begin(); q != validReadAddresses.end(); ++q)
      //      std::cout << *q << ' ';
      //DCOUT(endl);
      
    }
  }
};

#endif