[CTG] Replace set with OrderedSet (#570)

- Add new dependency 'ordered-set'
- Replace 'set' with 'OrderedSet', riscv-ctg use set to remove duplicates, but this
  will cause the unstable ordering of the for loop, and finally the generated
  tests will be different each time.

Co-authored-by: Umer Shahid <[email protected]>

requirements.txt

@@ -17,6 +17,7 @@ Jinja2
 m2r2>=0.2.7
 MarkupSafe>=1.1.1
 mistune>=0.8.4
+ordered-set>=4.1.0
 oyaml>=0.9
 packaging>=19.0
 pbr>=5.3.1

riscv-ctg/riscv_ctg/constants.py

@@ -3,6 +3,8 @@
 import os
 from math import *
 from string import Template
+
+from ordered_set import OrderedSet
 from riscv_isac.fp_dataset import *
 
 root = os.path.abspath(os.path.dirname(__file__))
@@ -120,7 +122,7 @@ def gen_sign_dataset(bit_width):
     t1 =( '' if bit_width%2 == 0 else '1') + ''.join(['01']*int(bit_width/2))
     t2 =( '' if bit_width%2 == 0 else '0') + ''.join(['10']*int(bit_width/2))
     data += [twos(t1,bit_width),twos(t2,bit_width)]
-    return list(set(data))
+    return list(OrderedSet(data))
 
 def gen_usign_dataset(bit_width):
     '''

riscv-ctg/riscv_ctg/cross_comb.py

@@ -122,7 +122,7 @@ def cross_comb(self, cgf_node):
         full_solution = []
 
         if 'cross_comb' in cgf_node:
-            cross_comb = set(cgf_node['cross_comb'])
+            cross_comb = OrderedSet(cgf_node['cross_comb'])
         else:
             return
 
@@ -210,7 +210,7 @@ def eval_conds(*oprs_lst):
                         opr_lst += get_oprs(assgn)
 
                         # Remove redundant operands
-                        opr_lst = list(set(opr_lst))
+                        opr_lst = list(OrderedSet(opr_lst))
 
                         # Get possible instructions
                         problem.reset()
@@ -293,7 +293,7 @@ def exc_rd_zero(*oprs_lst):
                     opr_lst = get_oprs(cond)
                     opr_lst += get_oprs(assgn)
 
-                    opr_lst = list(set(opr_lst))                  
+                    opr_lst = list(OrderedSet(opr_lst))                  
 
                     if data[i] in self.OP_TEMPLATE:                                    # If single instruction
                         instr = data[i]
@@ -363,7 +363,7 @@ def exc_rd_zero(*oprs_lst):
 
             full_solution += [solution]
 
-        self.isa = list(set(isa_set))
+        self.isa = list(OrderedSet(isa_set))
         return full_solution
 
     def swreg(cross_comb_instrs):
@@ -381,10 +381,10 @@ def swreg(cross_comb_instrs):
                 if key != 'instr' and key != 'imm_val':
                     op_vals.add(val)
 
-        swreg_sol = set(['x'+str(x) for x in range(0,32 if 'e' not in base_isa else 16)]) - op_vals
+        swreg_sol = OrderedSet(['x'+str(x) for x in range(0,32 if 'e' not in base_isa else 16)]) - op_vals
 
         sreg = random.choice(list(swreg_sol))
-        freg_Sol = swreg_sol - set(sreg)
+        freg_Sol = swreg_sol - OrderedSet(sreg)
         freg = random.choice(list(freg_Sol))
         return (sreg, freg)
 
@@ -401,7 +401,7 @@ def get_reginit_str(cross_comb_instrs, freg):
             - List of initialization strings
         '''
 
-        reg_init_lst = set()
+        reg_init_lst = OrderedSet()
 
         for instr_dict in cross_comb_instrs:
             if 'rd' in instr_dict:
@@ -455,7 +455,7 @@ def write_test(self, fprefix, cgf_node, usage_str, cov_label, full_solution):
             sig_label = "signature_" + sreg + "_" + str(sreg_dict[sreg])
             code = code + "\nRVTEST_SIGBASE(" + sreg + ", "+ sig_label + ")\n\n"
 
-            rd_lst = set()
+            rd_lst = OrderedSet()
             # Generate instruction corresponding to each instruction dictionary
             # Append signature update statements to store rd value after each instruction
             code += '// Cross-combination test sequence\n'

riscv-ctg/riscv_ctg/csr_comb.py

@@ -267,7 +267,7 @@ def __init__(self, base_isa, xlen, randomize):
     def csr_comb(self, cgf_node):
         logger.debug('Generating tests for csr_comb')
         if 'csr_comb' in cgf_node:
-            csr_comb = set(cgf_node['csr_comb'])
+            csr_comb = OrderedSet(cgf_node['csr_comb'])
         else:
             return
 

riscv-ctg/riscv_ctg/generator.py

@@ -340,25 +340,17 @@ def opcomb(self, cgf):
         solutions = []
         op_conds = {}
         opcomb_value = cgf.get("op_comb")
-        if "op_comb" in cgf:
-            op_comb = set(cgf["op_comb"])
-        else:
-            op_comb = set([])
+        op_comb = OrderedSet(opcomb_value or [])
         for op in self.op_vars:
-            if op in cgf:
-                op_conds[op] = set(cgf[op])
-            else:
-                op_conds[op] = set([])
+            op_conds[op] = OrderedSet(cgf.get(op, []))
         individual = False
         nodiff = False
         construct_constraint = lambda val: (lambda x: bool(x in val))
         while any([len(op_conds[x])!=0 for x in op_conds]+[len(op_comb)!=0]):
             cond_str = ''
             cond_vars = []
-            if self.random:
-                problem = Problem(MinConflictsSolver())
-            else:
-                problem = Problem()
+            solver = MinConflictsSolver() if self.random else None
+            problem = Problem(solver)
 
             done = False
             for var in self.op_vars:
@@ -418,7 +410,7 @@ def eval_func(cond):
                 for var,val in zip(self.op_vars,op_tuple):
                     locals()[var] = val
                 return eval(cond)
-            sat_set = set(filter(eval_func,op_comb))
+            sat_set = OrderedSet(filter(eval_func,op_comb))
             cond_str += ", ".join([var+"=="+solution[var] for var in cond_vars]+list(sat_set))
             op_tuple.append(cond_str)
             problem.reset()
@@ -450,7 +442,7 @@ def valcomb(self, cgf):
         val_comb = []
 
         conds = list(cgf['val_comb'].keys())
-        inds = set(range(len(conds)))
+        inds = OrderedSet(range(len(conds)))
         merge = True
         if 'fcvt' in self.opcode or 'fmv' in self.opcode:
             if self.opcode.split(".")[-1] in ['x','w','wu','l','lu']:
@@ -479,7 +471,7 @@ def valcomb(self, cgf):
                                 else:
                                     logger.error("Invalid Coverpoint: More than one value of "+ i +" found!")
                                     sys.exit(1)
-                if(set(d.keys()) != set(self.val_vars)):
+                if(OrderedSet(d.keys()) != OrderedSet(self.val_vars)):
                     logger.warning(
                         "Valcomb skip: Cannot bypass SAT Solver for partially defined coverpoints!"\
                                 + str(req_val_comb))
@@ -525,7 +517,7 @@ def eval_func(cond):
                     for var,val in zip(self.val_vars,val_tuple):
                         locals()[var] = val
                     return eval(cond)
-                sat_set=set(filter(lambda x: eval_func(conds[x]),inds))
+                sat_set=OrderedSet(filter(lambda x: eval_func(conds[x]),inds))
                 inds = inds - sat_set
                 val_tuple.append(req_val_comb+', '+', '.join([conds[i] for i in sat_set]))
                 problem.reset()
@@ -803,7 +795,7 @@ def gen_inst(self,op_comb, val_comb, cgf):
 
         for op,val_soln in zip(op_comb,val_comb):
             val = [x for x in val_soln]
-            if any([x=='x0' for x in op]) or not (len(op) == len(set(op))):
+            if any([x=='x0' for x in op]) or not (len(op) == len(OrderedSet(op))):
                 cont.append(val_soln)
                 op_inds = list(ind_dict.keys())
                 for i,x in enumerate(op_inds):
@@ -853,7 +845,7 @@ def gen_inst(self,op_comb, val_comb, cgf):
             else:
                 instr_dict.append(self.__instr__(op,val))
 
-        hits = defaultdict(lambda:set([]))
+        hits = defaultdict(lambda:OrderedSet([]))
         final_instr = []
 
         rm_dict = {
@@ -898,29 +890,29 @@ def eval_inst_coverage(coverpoints,instr):
                 var_dict.update(ext_specific_vars)
 
             if 'val_comb' in coverpoints:
-                valcomb_hits = set([])
+                valcomb_hits = OrderedSet([])
                 for coverpoint in coverpoints['val_comb']:
                     if eval(coverpoint,globals(),var_dict):
                         valcomb_hits.add(coverpoint)
                 cover_hits['val_comb']=valcomb_hits
             if 'op_comb' in coverpoints:
-                opcomb_hits = set([])
+                opcomb_hits = OrderedSet([])
                 for coverpoint in coverpoints['op_comb']:
                     if eval(coverpoint,globals(),var_dict):
                         opcomb_hits.add(coverpoint)
                 cover_hits['op_comb']=opcomb_hits
             if 'rs1' in coverpoints:
                 if var_dict['rs1'] in coverpoints['rs1']:
-                    cover_hits['rs1'] = set([var_dict['rs1']])
+                    cover_hits['rs1'] = OrderedSet([var_dict['rs1']])
             if 'rs2' in coverpoints:
                 if var_dict['rs2'] in coverpoints['rs2']:
-                    cover_hits['rs2'] = set([var_dict['rs2']])
+                    cover_hits['rs2'] = OrderedSet([var_dict['rs2']])
             if 'rs3' in coverpoints:
                 if var_dict['rs3'] in coverpoints['rs3']:
-                    cover_hits['rs3'] = set([var_dict['rs3']])
+                    cover_hits['rs3'] = OrderedSet([var_dict['rs3']])
             if 'rd' in coverpoints:
                 if var_dict['rd'] in coverpoints['rd']:
-                    cover_hits['rd'] = set([var_dict['rd']])
+                    cover_hits['rd'] = OrderedSet([var_dict['rd']])
             return cover_hits
         i = 0
 

riscv-isac/riscv_isac/coverage.py

@@ -188,7 +188,7 @@ def __init__(self,label,coverpoint,xlen,flen,addr_pairs,sig_addrs,window_size,in
         self.result = 0
         self.queue = []
 
-        self.tracked_regs = set()
+        self.tracked_regs = OrderedSet()
         self.instr_addr_of_tracked_reg = {} # tracked_reg: instr_addr of instr which triggered its tracking
         self.instr_stat_meta_at_addr = {} # start_instr_addr: [is_ucovpt, num_exp, num_obs, num_rem, covpts_hit, code_seq, store_addresses, store_vals]
 
@@ -253,7 +253,7 @@ def compute_cross_cov(self):
         Also perform tracking for generating the data propagation report
         '''
         hit_covpt = False
-        regs_to_track = set()
+        regs_to_track = OrderedSet()
 
         for index in range(len(self.ops)):
             instr = self.queue[index]
@@ -920,8 +920,8 @@ def compute_per_line(queue, event, cgf_queue, stats_queue, cgf, xlen, flen, addr
     inxFlg = arch_state.inxFlg
 
     # Set of elements to monitor for tracking signature updates
-    tracked_regs_immutable = set()
-    tracked_regs_mutable = set()
+    tracked_regs_immutable = OrderedSet()
+    tracked_regs_mutable = OrderedSet()
     tracked_instrs = [] # list of tuples of the type (list_instr_names, triggering_instr_addr)
 
     instr_stat_meta_at_addr = {} # Maps an address to the stat metadata of the instruction present at that address [is_ucovpt, num_exp, num_obs, num_rem, covpts_hit, code_seq, store_addresses, store_vals]
@@ -1262,7 +1262,7 @@ def get_key_from_value(dictionary, target_value):
                 if hit_csr_covpt:
                     stats.cov_pt_sig += covpt
 
-                    csr_regs_involved_in_covpt = set()
+                    csr_regs_involved_in_covpt = OrderedSet()
                     for covpt in csr_covpt:
                         for reg in csr_reg_num_to_str.values():
                             if reg in covpt:
@@ -1717,7 +1717,7 @@ def compute(trace_file, test_name, cgf, parser_name, decoder_name, detailed, xle
                 _x = (hex(x[0]), hex(x[1]), str(int((x[1]-x[0])/4)) + ' words')
             sig_addrs_hex.append(_x)
 
-        cov_set = set()
+        cov_set = OrderedSet()
         count = 1
         for addrs,vals,cover,code in stats.stat1:
             sig = ''
@@ -1753,7 +1753,7 @@ def compute(trace_file, test_name, cgf, parser_name, decoder_name, detailed, xle
             stat3_log += _l + '\n\n'
 
         stat5_log = ''
-        sig_set = set()
+        sig_set = OrderedSet()
         overwrites = 0
         for addr, val, cover, code in stats.stat5:
             if addr in sig_set:

riscv-isac/riscv_isac/data/rvopcodesdecoder.py

@@ -3,6 +3,7 @@
 from collections import defaultdict
 import pprint
 import os
+from ordered_set import OrderedSet
 
 from constants import *
 #from riscv_isac.data.constants import *
@@ -232,13 +233,13 @@ def build_instr_dict(inst_dict):
         funct_occ = [funct[0] for ins in funct_list for funct in ins]
 
         # Path recoder
-        funct_path = set()
+        funct_path = OrderedSet()
         # Check if there are functions remaining
         while funct_occ:
             if (1, 0) in funct_occ:
                 max_funct = (1, 0)
             else:
-                max_funct = max(set(funct_occ),key=funct_occ.count)
+                max_funct = max(OrderedSet(funct_occ),key=funct_occ.count)
 
             funct_occ = list(filter(lambda a: a != max_funct, funct_occ))
 

riscv-isac/riscv_isac/fp_dataset.py

@@ -4,6 +4,7 @@
 import random
 import sys
 import math
+from ordered_set import OrderedSet
 from decimal import *
 
 # Prasanna
@@ -4941,7 +4942,7 @@ def ibm_b24(flen, iflen, opcode, ops, inxFlg=False):
         t = "{:e}".format(data[0])
         b24_comb.append((floatingPoint_tohex(iflen,float(t)),data[1]))
 
-    b24_comb = set(b24_comb)
+    b24_comb = OrderedSet(b24_comb)
 
     coverpoints = []
     k=0

riscv-isac/riscv_isac/plugins/internaldecoder.py

@@ -1,3 +1,4 @@
+from ordered_set import OrderedSet
 import riscv_isac.plugins as plugins
 
 class disassembler():
@@ -38,9 +39,9 @@ def __init__(self):
         self.C_OPCODES = C_OPCODES
         self.OPCODES = OPCODES
         self.init_rvp_dictionary()
-        self.rvp_rs1_is_64bit_set = set('smal add64 radd64 uradd64 kadd64 ukadd64 sub64 rsub64 ursub64 ksub64 uksub64 wext wexti'.split())
-        self.rvp_rs2_is_64bit_set = set(     'add64 radd64 uradd64 kadd64 ukadd64 sub64 rsub64 ursub64 ksub64 uksub64'.split())
-        self.rvp_rd_is_64bit_set  = set('smul16 smulx16 umul16 umulx16 smul8 smulx8 umul8 umulx8 smal add64 radd64 uradd64 kadd64 ukadd64 sub64 rsub64 ursub64 ksub64 uksub64 smar64 smsr64 umar64 umsr64 kmar64 kmsr64 ukmar64 ukmsr64 smalbb smalbt smaltt smalda smalxda smalds smaldrs smalxds smslda smslxda mulr64 mulsr64 wext wexti'.split())
+        self.rvp_rs1_is_64bit_set = OrderedSet('smal add64 radd64 uradd64 kadd64 ukadd64 sub64 rsub64 ursub64 ksub64 uksub64 wext wexti'.split())
+        self.rvp_rs2_is_64bit_set = OrderedSet(     'add64 radd64 uradd64 kadd64 ukadd64 sub64 rsub64 ursub64 ksub64 uksub64'.split())
+        self.rvp_rd_is_64bit_set  = OrderedSet('smul16 smulx16 umul16 umulx16 smul8 smulx8 umul8 umulx8 smal add64 radd64 uradd64 kadd64 ukadd64 sub64 rsub64 ursub64 ksub64 uksub64 smar64 smsr64 umar64 umsr64 kmar64 kmsr64 ukmar64 ukmsr64 smalbb smalbt smaltt smalda smalxda smalds smaldrs smalxds smslda smslxda mulr64 mulsr64 wext wexti'.split())
 
     def init_rvp_dictionary(self):
         # Create RVP Dictiory 0 for instruction:  clrs8  clrs16  clrs32  clo8  clo16  clo32  clz8  clz16  clz32  kabs8  kabs16  kabsw  sunpkd810  sunpkd820  sunpkd830  sunpkd831  sunpkd832  swap8  zunpkd810  zunpkd820  zunpkd830  zunpkd831  zunpkd832  kabs32