diff --git a/wally-pipelined/testgen/testgen-ADD-SUB.py b/wally-pipelined/testgen/testgen-ADD-SUB-SLT-SLTU-XOR-OR-AND.py similarity index 82% rename from wally-pipelined/testgen/testgen-ADD-SUB.py rename to wally-pipelined/testgen/testgen-ADD-SUB-SLT-SLTU-XOR-OR-AND.py index 16e02be6..9f795036 100755 --- a/wally-pipelined/testgen/testgen-ADD-SUB.py +++ b/wally-pipelined/testgen/testgen-ADD-SUB-SLT-SLTU-XOR-OR-AND.py @@ -1,6 +1,6 @@ #!/usr/bin/python3 ################################## -# testgen-ADD-SUB.py +# testgen-ADD-SUB-SLT-SLTU-XOR-OR-AND.py # # David_Harris@hmc.edu 19 January 2021 # @@ -19,11 +19,34 @@ from random import getrandbits # functions ################################## -def computeExpected(a, b, test): +def twoscomp(a): + amsb = a >> (xlen-1) + alsbs = ((1 << (xlen-1)) - 1) & a + if (amsb): + asigned = a - (1<> (xlen-1) + alsbs = ((1 << (xlen-1)) - 1) & a + if (amsb): + asigned = a - (1<> b + elif (test == "SRA"): + return asigned >> b + else: + die("bad test name ", test) + # exit(1) + +def randRegs(): + reg1 = randint(1,31) + reg2 = randint(1,31) + reg3 = randint(1,31) + if (reg1 == 6 or reg2 == 6 or reg3 == 6 or reg1 == reg2): + return randRegs() + else: + return reg1, reg2, reg3 + +def writeVector(a, b, storecmd, xlen): + global testnum + expected = computeExpected(a, b, test, xlen) + expected = expected % 2**xlen # drop carry if necessary + if (expected < 0): # take twos complement + expected = 2**xlen + expected + reg1, reg2, reg3 = randRegs() + lines = "\n# Testcase " + str(testnum) + ": rs1:x" + str(reg1) + "(" + formatstr.format(a) + lines = lines + "), rs2:x" + str(reg2) + "(" +formatstr.format(b) + lines = lines + "), result rd:x" + str(reg3) + "(" + formatstr.format(expected) +")\n" + lines = lines + "li x" + str(reg1) + ", MASK_XLEN(" + formatstr.format(a) + ")\n" + lines = lines + "li x" + str(reg2) + ", MASK_XLEN(" + formatstr.format(b) + ")\n" + lines = lines + test + " x" + str(reg3) + ", x" + str(reg1) + ", x" + str(reg2) + "\n" + lines = lines + storecmd + " x" + str(reg3) + ", " + str(wordsize*testnum) + "(x6)\n" + lines = lines + "RVTEST_IO_ASSERT_GPR_EQ(x7, " + str(reg3) +", "+formatstr.format(expected)+")\n" + f.write(lines) + if (xlen == 32): + line = formatrefstr.format(expected)+"\n" + else: + line = formatrefstr.format(expected % 2**32)+"\n" + formatrefstr.format(expected >> 32) + "\n" + r.write(line) + testnum = testnum+1 + +################################## +# main body +################################## + +# change these to suite your tests +tests = ["SLL", "SRL", "SRA"] +author = "David_Harris@hmc.edu" +xlens = [32, 64] +numrand = 48 + +# setup +seed(0) # make tests reproducible + +# generate files for each test +for xlen in xlens: + formatstrlen = str(int(xlen/4)) + formatstr = "0x{:0" + formatstrlen + "x}" # format as xlen-bit hexadecimal number + formatrefstr = "{:08x}" # format as xlen-bit hexadecimal number with no leading 0x + if (xlen == 32): + storecmd = "sw" + wordsize = 4 + else: + storecmd = "sd" + wordsize = 8 + for test in tests: + corners = [0, 1, 2, 0xFF, 0x624B3E976C52DD14 % 2**xlen, 2**(xlen-1)-2, 2**(xlen-1)-1, + 2**(xlen-1), 2**(xlen-1)+1, 0xC365DDEB9173AB42 % 2**xlen, 2**(xlen)-2, 2**(xlen)-1] + if (xlen == 32): + shamt = [0, 1, 2, 3, 4, 8, 15, 16, 29, 30, 31] + else: + shamt = [0, 1, 3, 8, 15, 16, 29, 31, 32, 47, 48, 62, 63] + + imperaspath = "../../imperas-riscv-tests/riscv-test-suite/rv" + str(xlen) + "i/" + basename = "WALLY-" + test + fname = imperaspath + "src/" + basename + ".S" + refname = imperaspath + "references/" + basename + ".reference_output" + testnum = 0 + + # print custom header part + f = open(fname, "w") + r = open(refname, "w") + line = "///////////////////////////////////////////\n" + f.write(line) + lines="// "+fname+ "\n// " + author + "\n" + f.write(lines) + line ="// Created " + str(datetime.now()) + f.write(line) + + # insert generic header + h = open("testgen_header.S", "r") + for line in h: + f.write(line) + + # print directed and random test vectors + for a in corners: + for b in shamt: + writeVector(a, b, storecmd, xlen) + for i in range(0,numrand): + a = getrandbits(xlen) + b = getrandbits(xlen) + writeVector(a, b, storecmd, xlen) + + + # print footer + h = open("testgen_footer.S", "r") + for line in h: + f.write(line) + + # Finish + lines = ".fill " + str(testnum) + ", " + str(wordsize) + ", -1\n" + lines = lines + "\nRV_COMPLIANCE_DATA_END\n" + f.write(lines) + f.close() + r.close() + + + +