mutiple fixes: parallel printf, fixed cycle in cache, opencl refactored vecadd and sgemm, regen opencl kernels with hard-float, fixed vortex io bus interface, fixed dpi floats APi to support multicore mode, make vlsim multicore default, make rtlsim multi-core default, removed POCL binaries from repository, updated Makefiles to use external POCL

hw/rtl/fp_cores/svdpi/float_dpi.cpp

@@ -3,11 +3,13 @@
 #include <unordered_map>
 #include <vector>
 #include <mutex>
+#include <iostream>
 #include "svdpi.h"
 #include "verilated_vpi.h"
 #include "VX_config.h"
 
 extern "C" {
+  int dpi_register();
   void dpi_fadd(int inst, bool enable, int a, int b, int* result);
   void dpi_fsub(int inst, bool enable, int a, int b, int* result);
   void dpi_fmul(int inst, bool enable, int a, int b, int* result);
@@ -66,19 +68,28 @@ union Float_t {
 class Instances {
 public:
   ShiftRegister& get(int inst) {
-    mutex_.lock();
-    ShiftRegister& sr = instances_[inst];
+    return instances_.at(inst);
+  }
+
+  int allocate() {
+    mutex_.lock();   
+    int inst = instances_.size();
+    instances_.resize(inst + 1); 
     mutex_.unlock();
-    return sr;
+    return inst;
   }
 
 private:
-  std::unordered_map<int, ShiftRegister> instances_;
+  std::vector<ShiftRegister> instances_;
   std::mutex mutex_;
 };
 
 Instances instances;
 
+int dpi_register() {
+  return instances.allocate();
+}
+
 void dpi_fadd(int inst, bool enable, int a, int b, int* result) {
   ShiftRegister& sr = instances.get(inst);
 

hw/rtl/fp_cores/svdpi/float_dpi.vh

@@ -1,6 +1,8 @@
 `ifndef FLOAT_DPI
 `define FLOAT_DPI
 
+import "DPI-C" context function int dpi_register();
+
 import "DPI-C" context function void dpi_fadd(int inst, input logic enable, input int a, input int b, output int result);
 import "DPI-C" context function void dpi_fsub(int inst, input logic enable, input int a, input int b, output int result);
 import "DPI-C" context function void dpi_fmul(int inst, input logic enable, input int a, input int b, output int result);