From: Alexandre Oliva Date: Sun, 29 Nov 2020 15:52:00 +0000 (-0300) Subject: comp16-v1-skel: first cut at the logic for attempt 1 encoding, untested X-Git-Tag: convert-csv-opcode-to-binary~1627 X-Git-Url: https://git.libre-soc.org/?a=commitdiff_plain;h=10ea2b5370c598630a1d30fb8b496b7a7d2d0925;p=libreriscv.git comp16-v1-skel: first cut at the logic for attempt 1 encoding, untested --- diff --git a/lxo/532/comp16-v1-skel.py b/lxo/532/comp16-v1-skel.py new file mode 100644 index 000000000..d4a685205 --- /dev/null +++ b/lxo/532/comp16-v1-skel.py @@ -0,0 +1,260 @@ +#! /bin/env python3 + +# Feed this script the output of objdump -M raw --no-show-raw-insn ppc-prog + +# It will look for insns that can be represented in compressed mode, +# according to the encoding rules in the copcond dictionary below. + +# Nothing is assumed as to the actual bit-encoding of the insns, this +# is just to experiment with insn selection and get a quick feedback +# loop for the encoding options in compressed mode. + +# In this script, the computations of encoding modes and transitions +# are those for attempt 1 encoding, that encompasses: + +# - a 16-bit insn (with 10-bit payload) that may switch to compressed +# mode or return to 32-bit mode; + +# - 16-bit insns in compressed mode, each with 2 bits devoted to +# encoding one of the following possibilities: + +# -- switch back to uncompressed mode at the next insn + +# -- interpret the next insn in uncompressed mode, then return to +# compressed mode + +# -- remain in 16-bit mode for the next insn + +# -- take the 16 bits that would be the next compressed insn as an +# extension to the present 16-bit insn, and remain in 16-bit mode for +# the subsequent 16-bits + +# At (visible) entry points, mode is forced to return to uncompressed +# mode. + +# The entire code stream is printed, without any attempt to modify the +# addresses that go along with or in them; we only insert markers for +# the transition points, and for the compressed instructions. + +# The really useful information is printed at the end: a summary of +# transition and compressed-insn counts, and the achieved compression +# rate. + +import sys +import re + +insn = re.compile('\s+(?P[0-9a-f]+):\s+(?P[^ ]+) *(?P.*)') + +opkind = re.compile('(?P(?P[cf]?r)(?P[0-9]+))|(?P-?[0-9]+)|(?P[0-9a-f]+)(?: <.*>)?|(?P-?[0-9]+)\((?Pr[0-9]+)\)') + +def mapop(op): + match = opkind.fullmatch(op) + + if match is None: + op = ('other', op) + elif match['reg'] is not None: + op = (match['regkind'], int(match['regnum']), op) + elif match['immediate'] is not None: + op = ('imm', int (op).bit_length (), op) + elif match['branch'] is not None: + op = ('pcoff', (int (match['branch'], 16) + - int (addr, 16)).bit_length (), op, addr) + elif match['offset'] is not None: + op = ('ofst', mapop(match['offset']), mapop(match['basereg']), op) + else: + raise "unrecognized operand kind" + + return op + +def opclass(mop): + return mop[0] +def regno(mop): + if mop[0] in { 'r', 'fr', 'cr' }: + return mop[1] + else: + raise "operand is not a register" + +def immbits(mop): + if mop[0] is 'imm': + return mop[1] + else: + raise "operand is not an immediate" + +# Following are predicates to be used in copcond, to tell the mode in +# which opcode with ops as operands is to be represented + +# Any occurrence of the opcode can be compressed. +def anyops(opcode, ops): + return 1 + +# Compress iff first and second operands are the same. +def same01(opcode, ops): + if ops[0] == ops[1]: + return 1 + else: + return 0 + +# Registers representable in a made-up 3-bit mapping. +cregs2 = { 1, 2, 3, 4, 5, 6, 7, 31 } + +# Return true iff mop is a regular register present in cregs2 +def bin2regs3(mop): + return opclass(mop) is 'r' and regno(mop) in cregs2 + +# Return true iff mop is an immediate of at most 8 bits. +def imm8(mop): + return opclass(mop) is 'imm' and immbits(mop) <= 8 + +# Compress binary opcodes iff the first two operands (output and first +# input operand) are registers representable in 3 bits in compressed +# mode, and the immediate operand can be represented in 8 bits. +def bin2regs3imm8(opcode, ops): + if bin2regs3(ops[0]) and bin2regs3(ops[1]) and imm8(ops[2]): + return 1 + else: + return 0 + +# Map opcodes that might be compressed to a function that returns the +# best potential encoding kind for the insn, per the numeric coding +# below. +copcond = { + +} + +# We have 4 kinds of insns: + +# 0: uncompressed; leave input insn unchanged +# 1: 16-bit compressed, only in compressed mode +# 2: 16-bit extended by another 16-bit, only in compressed mode +# 3: 10-bit compressed, may switch to compressed mode + +# count[0:3] count the occurrences of the base kinds. +# count[4] counts extra 10-bit nop-switches to compressed mode, +# tentatively introduced before insns that can be 16-bit encoded. +count = [0,0,0,0,0] +# Default comments for the insn kinds above. 2 is always tentative. +comments = ['', '\t; 16-bit', '\t; tentative 16+16-bit', '\t; 10-bit'] + +# cur stands for the insn kind that we read and processed in the +# previous iteration of the loop, and prev is the one before it. the +# one we're processing in the current iteration will be stored in +# next until we make it cur at the very end of the loop. +prev = cur = 0 + +for line in sys.stdin: + if line[-1] is '\n': + line = line[:-1] + + match = insn.fullmatch(line) + if match is None: + print(line) + # Switch to uncompressed mode at function boundaries + prev = prev2 = 0 + continue + + addr = match['addr'] + opcode = match['opcode'] + operands = match['operands'] + + if opcode in copcond: + next = copcond[opcode](opcode, + [mapop(op) for op in operands.split(',')]) + else: + next = 0 + + comment = None + + if cur is 0: + if next is 0: + True # Uncompressed mode for good. + elif next is 1: + # If cur was not a single uncompressed mode insn, + # tentatively encode a 10-bit nop to enter compressed + # mode, and then 16-bit. It takes as much space as + # encoding as 32-bit, but offers more possibilities for + # subsequent compressed encodings. A compressor proper + # would have to go back and change the encoding + # afterwards, but wé re just counting. + if prev is not 1: + print('\t\th.nop\t\t; tentatively switch to compressed mode') + count[4] += 1 + comment = 'tentatively compressed to 16-bit' + elif next is 2: + # We can use compressed encoding for next after an + # uncompressed insn only if it's the single-insn + # uncompressed mode slot. For anything else, we're better + # off using uncompressed encoding for next, since it makes + # no sense to spend a 10-bit nop to switch to compressed + # mode for a 16+16-bit insn. If subsequent insns would + # benefit from compressed encoding, we can switch then. + if prev is not 1: + next = 0 + comment = 'not worth a nop for 16+16-bit' + elif next is 3: + # If prev was 16-bit compressed, cur would be in the + # single-insn uncompressed slot, so next could be encoded + # as 16-bit, enabling another 1-insn uncompressed slot + # after next that a 10-bit insn wouldn't, so make it so. + if prev is 1: + next = 1 + comment = '16-bit, could be 10-bit' + elif cur is 1: + # After a 16-bit insn, anything goes. If it remains in 16-bit + # mode, we can have 1 or 2 as next; if it returns to 32-bit + # mode, we can have 0 or 3. Using 1 instead of 3 makes room + # for a subsequent single-insn compressed mode, so prefer + # that. + if next is 3: + next = 1 + comment = '16-bit, could be 10-bit' + elif cur is 2: + # After a 16+16-bit insn, we can't switch directly to 32-bit + # mode. However, cur could have been encoded as 32-bit, since + # any 16+16-bit insn can. Indeed, we may have an arbitrary + # long sequence of 16+16-bit insns before next, and if next + # can only be encoded in 32-bit mode, we can "resolve" all + # previous adjacent 16+16-bit insns to the corresponding + # 32-bit insns in the encoding, and "adjust" the 16-bit or + # 10-bit insn that enabled the potential 16+16-bit encoding to + # switch to 32-bit mode then instead. + if next is 0: + prev = cur = 0 + comment = '32-bit, like tentative 16+16-bit insns above' + elif cur is 3: + # After a 10-bit insn, another insn that could be encoded as + # 10-bit might as well be encoded as 16-bit, to make room for + # a single-insn uncompressed insn afterwards. + if next is 3: + next = 1 + comment = '16-bit, could be 10-bit' + else: + raise "unknown mode for previous insn" + + count[next] += 1 + + if comment is None: + comment = comments[next] + else: + comment = '\t; ' + comment + + print(line + comment) + + prev = cur + cur = next + +transition_bytes = 2 * count[4] +compressed_bytes = 2 * (count[1] + count[3]) +uncompressed_bytes = 4 * (count[0] + count[2]) +total_bytes = transition_bytes + compressed_bytes + uncompressed_bytes +original_bytes = 2 * compressed_bytes + uncompressed_bytes + +print() +print('Summary') +print('32-bit uncompressed instructions: %i' % count[0]) +print('16-bit compressed instructions: %i' % count[1]) +print('16+16-bit (tentative) compressed-mode instructions: %i' % count[2]) +print('10-bit compressed instructions: %i' % count[3]) +print('10-bit (tentative) mode-switching nops: %i' % count[4]) +print('Compressed size estimate: %i' % total_bytes) +print('Original size: %i' % original_bytes) +print('Compressed/original ratio: %f' % (total_bytes / original_bytes)) diff --git a/lxo/ChangeLog b/lxo/ChangeLog index 6db83291d..adb88a1d2 100644 --- a/lxo/ChangeLog +++ b/lxo/ChangeLog @@ -1,5 +1,25 @@ 2020-11-28 + * 532: Figured out and implemented the logic to infer mode + switching for best compression under attempt 1 proposed encoding, + namely with 10-bit insns, 16-bit insns, 16+16-bit insns, and + 32-bit insns. 10-bit insns appear in uncompressed mode, and can + be followed by insns in either mode; 16-bit ones appear in + compressed mode, and can remain in compressed mode, or switch to + uncomprssed mode for 1 insn or for good; 16+16-bit ones appear in + compressed mode, and cannot switch modes; 32-bit ones appear only + in uncompressed mode, or in the single-insn slot after a 16-bit + that requests it. If we find a 16-bit insn while we're in + uncompressed mode, use a 10-bit nop to tentatively switch. Insns + that can be encoded in 10-bits, but appear in compressed mode, had + better be encoded in 16-bits, for that offers further subsequent + encoding options, without downsides for size estimation. Insns + that can be encoded as 16+16-bit decay to 32-bit if in + uncompressed mode, or if, after a sequence thereof, a later insn + forces a switch to 32-bit mode without an intervening switching + insn. Still missing: the code to select what insns can be encoded + in what modes. (6:42) + * 532: Implemented a skeleton for compression ratio estimation, initially with the simpler mode switching of the 8-bit nop, odd-address 16-bit insns. Next, rewrite it for all the complexity