Mesa (main): pan/va: Add Valhall assembler

GitLab Mirror gitlab-mirror at kemper.freedesktop.org
Tue Jul 27 20:33:38 UTC 2021


Module: Mesa
Branch: main
Commit: 508dfba9137347baa34c99ea6a58c05eb0dc27d9
URL:    http://cgit.freedesktop.org/mesa/mesa/commit/?id=508dfba9137347baa34c99ea6a58c05eb0dc27d9

Author: Alyssa Rosenzweig <alyssa at collabora.com>
Date:   Fri Jul 16 10:52:43 2021 -0400

pan/va: Add Valhall assembler

This Python script acts as a standalone assembler. It takes Valhall
instructions with Mesa-flavour syntax, parses them, errors out if there
are syntax errors, and writes out an assembled binary if there are not.
It also is available as a programmatic interface for automated testing.

While this attempts to handle syntax errors, it does not check
semantics. It will happily compile programs that fault, provided each
instruction locally 'looks' plausible.

The code itself is quite small, despite supporting most of the known
ISA, because the syntax is regular and the heavylifting is done by
ISA.xml and valhall.py.

Signed-off-by: Alyssa Rosenzweig <alyssa at collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12025>

---

 src/panfrost/bifrost/valhall/asm.py | 349 ++++++++++++++++++++++++++++++++++++
 1 file changed, 349 insertions(+)

diff --git a/src/panfrost/bifrost/valhall/asm.py b/src/panfrost/bifrost/valhall/asm.py
new file mode 100644
index 00000000000..2a4702a5702
--- /dev/null
+++ b/src/panfrost/bifrost/valhall/asm.py
@@ -0,0 +1,349 @@
+#encoding=utf-8
+
+# Copyright (C) 2021 Collabora, Ltd.
+#
+# Permission is hereby granted, free of charge, to any person obtaining a
+# copy of this software and associated documentation files (the "Software"),
+# to deal in the Software without restriction, including without limitation
+# the rights to use, copy, modify, merge, publish, distribute, sublicense,
+# and/or sell copies of the Software, and to permit persons to whom the
+# Software is furnished to do so, subject to the following conditions:
+#
+# The above copyright notice and this permission notice (including the next
+# paragraph) shall be included in all copies or substantial portions of the
+# Software.
+#
+# THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+# IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+# FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+# THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+# LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
+# FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
+# IN THE SOFTWARE.
+
+import argparse
+import sys
+import struct
+from valhall import instructions, enums, immediates, typesize
+
+LINE = ''
+
+class ParseError(Exception):
+    def __init__(self, error):
+        self.error = error
+
+# When running standalone, exit with the error since we're dealing with a
+# human. Otherwise raise a Python exception so the test harness can handle it.
+def die(s):
+    if __name__ == "__main__":
+        print(LINE)
+        print(s)
+        sys.exit(1)
+    else:
+        raise ParseError(s)
+
+def die_if(cond, s):
+    if cond:
+        die(s)
+
+def parse_int(s, minimum, maximum):
+    try:
+        number = int(s, base = 0)
+    except ValueError:
+        die(f"Expected number {s}")
+
+    if number > maximum or number < minimum:
+        die(f"Range error on {s}")
+
+    return number
+
+def encode_source(op, immediate_mode = 'none'):
+    if op == 'atest_datum':
+        die_if(immediate_mode != 'none', 'Expected no modifier with atest datum')
+        return 0x2A | 0xC0
+    elif op.startswith('blend_descriptor_'):
+        die_if(immediate_mode != 'none', 'Expected no modifier with blend descriptor')
+        fin = op[len('blend_descriptor_'):]
+        die_if(len(fin) != 3, 'Bad syntax')
+        die_if(fin[1] != '_', 'Bad syntax')
+        die_if(fin[2] not in ['x', 'y'], 'Bad component')
+
+        rt = parse_int(fin[0], 0, 7)
+        hi = 1 if (fin[2] == 'y') else 0
+        return (0x30 | (2*rt) + hi) | 0xC0
+    elif op[0] == '`':
+        die_if(op[1] != 'r', f"Expected register after discard {op}")
+        return parse_int(op[2:], 0, 63) | 0x40
+    elif op[0] == 'r':
+        return parse_int(op[1:], 0, 63)
+    elif op[0] == 'u':
+        return parse_int(op[1:], 0, 63) | 0x80
+    elif op[0] == 'i':
+        return int(op[3:]) | 0xC0
+    elif op in enums['thread_storage_pointers'].bare_values:
+        die_if(immediate_mode != 'ts',
+                f'Expected .ts with thread pointer, got {immediate_mode}')
+
+        idx = 32 + enums['thread_storage_pointers'].bare_values.index(op)
+        return idx | 0xC0
+    elif op in enums['thread_identification'].bare_values:
+        die_if(immediate_mode != 'id',
+                'Expected .id modifier with thread storage pointer')
+
+        idx = 32 + enums['thread_identification'].bare_values.index(op)
+        return idx | 0xC0
+    elif op.startswith('0x'):
+        try:
+            val = int(op, base=0)
+        except ValueError:
+            die('Expected value')
+
+        die_if(val not in immediates, 'Unexpected immediate value')
+        return immediates.index(val) | 0xC0
+    else:
+        die('Invalid operand')
+
+def encode_dest(op):
+    die_if(op[0] != 'r', f"Expected register destination {op}")
+
+    parts = op.split(".")
+    reg = parts[0]
+
+    # Default to writing in full
+    wrmask = 0x3
+
+    if len(parts) > 1:
+        WMASKS = ["h0", "h1"]
+        die_if(len(parts) > 2, "Too many modifiers")
+        mask = parts[1];
+        die_if(mask not in WMASKS, "Expected a write mask")
+        wrmask = 1 << WMASKS.index(mask)
+
+    return parse_int(reg[1:], 0, 63) | (wrmask << 6)
+
+def parse_asm(line):
+    global LINE
+    LINE = line # For better errors
+    encoded = 0
+
+    # Figure out mnemonic
+    head = line.split(" ")[0]
+    opts = [ins for ins in instructions if head.startswith(ins.name)]
+    opts = sorted(opts, key=lambda x: len(x.name), reverse=True)
+
+    if len(opts) == 0:
+        die(f"No known mnemonic for {head}")
+
+    if len(opts) > 1 and len(opts[0].name) == len(opts[1].name):
+        print(f"Ambiguous mnemonic for {head}")
+        print(f"Options:")
+        for ins in opts:
+            print(f"  {ins}")
+        sys.exit(1)
+
+    ins = opts[0]
+
+    # Split off modifiers
+    if len(head) > len(ins.name) and head[len(ins.name)] != '.':
+        die(f"Expected . after instruction in {head}")
+
+    mods = head[len(ins.name) + 1:].split(".")
+    modifier_map = {}
+    immediate_mode = 'none'
+
+    for mod in mods:
+        if mod in enums['immediate_mode'].bare_values:
+            die_if(immediate_mode != 'none', 'Multiple immediate modes specified')
+            immediate_mode = mod
+
+    tail = line[(len(head) + 1):]
+    operands = [x.strip() for x in tail.split(",") if len(x.strip()) > 0]
+    expected_op_count = len(ins.srcs) + len(ins.dests) + len(ins.immediates) + len(ins.staging)
+    if len(operands) != expected_op_count:
+        die(f"Wrong number of operands in {line}, expected {expected_op_count}, got {len(operands)} {operands}")
+
+    # Encode each operand
+    for i, (op, sr) in enumerate(zip(operands, ins.staging)):
+        die_if(op[0] != '@', f'Expected staging register, got {op}')
+        parts = op[1:].split(':')
+
+        die_if(any([x[0] != 'r' for x in parts]), f'Expected registers, got {op}')
+        regs = [parse_int(x[1:], 0, 63) for x in parts]
+
+        sr_count = len(regs)
+        die_if(sr_count < 1, f'Expected staging register, got {op}')
+        die_if(sr_count > 7, f'Too many staging registers {sr_count}')
+
+        base = regs[0]
+        die_if(any([reg != (base + i) for i, reg in enumerate(regs)]),
+                'Expected consecutive staging registers, got {op}')
+
+        if sr.count == 0:
+            modifier_map["staging_register_count"] = sr_count
+        else:
+            die_if(sr_count != sr.count, f"Expected 4 staging registers, got {sr_count}")
+
+        encoded |= ((sr.encoded_flags | base) << sr.start)
+    operands = operands[len(ins.staging):]
+
+    for op, dest in zip(operands, ins.dests):
+        encoded |= encode_dest(op) << 40
+    operands = operands[len(ins.dests):]
+
+    if len(ins.dests) == 0 and len(ins.staging) == 0:
+        # Set a placeholder writemask to prevent encoding faults
+        encoded |= (0xC0 << 40)
+
+    for i, (op, src) in enumerate(zip(operands, ins.srcs)):
+        parts = op.split('.')
+        encoded |= encode_source(parts[0], immediate_mode) << (i * 8)
+
+        # Has a swizzle been applied yet?
+        swizzled = False
+
+        for mod in parts[1:]:
+            # Encode the modifier
+            if mod in src.offset and src.bits[mod] == 1:
+                encoded |= (1 << src.offset[mod])
+            elif mod in enums[f'swizzles_{src.size}_bit'].bare_values and (src.widen or src.lanes):
+                die_if(swizzled, "Multiple swizzles specified")
+                swizzled = True
+                val = enums[f'swizzles_{src.size}_bit'].bare_values.index(mod)
+                encoded |= (val << src.offset['widen'])
+            elif src.lane and mod in enums[f'lane_{src.size}_bit'].bare_values:
+                die_if(swizzled, "Multiple swizzles specified")
+                swizzled = True
+                val = enums[f'lane_{src.size}_bit'].bare_values.index(mod)
+                encoded |= (val << src.offset['lane'])
+            elif src.size == 32 and mod in enums['widen'].bare_values:
+                die_if(not src.swizzle, "Instruction doesn't take widens")
+                die_if(swizzled, "Multiple swizzles specified")
+                swizzled = True
+                val = enums['widen'].bare_values.index(mod)
+                encoded |= (val << src.offset['swizzle'])
+            elif src.size == 16 and mod in enums['swizzles_16_bit'].bare_values:
+                die_if(not src.swizzle, "Instruction doesn't take swizzles")
+                die_if(swizzled, "Multiple swizzles specified")
+                swizzled = True
+                val = enums['swizzles_16_bit'].bare_values.index(mod)
+                encoded |= (val << src.offset['swizzle'])
+            elif mod in enums['lane_8_bit'].bare_values:
+                die_if(not src.lane, "Instruction doesn't take a lane")
+                die_if(swizzled, "Multiple swizzles specified")
+                swizzled = True
+                val = enums['lane_8_bit'].bare_values.index(mod)
+                encoded |= (val << src.lane)
+            else:
+                die(f"Unknown modifier {mod}")
+
+        # Encode the identity if a swizzle is required but not specified
+        if src.swizzle and not swizzled and src.size == 16:
+            mod = enums['swizzles_16_bit'].default
+            val = enums['swizzles_16_bit'].bare_values.index(mod)
+            encoded |= (val << src.offset['swizzle'])
+        elif src.widen and not swizzled and src.size == 16:
+            die_if(swizzled, "Multiple swizzles specified")
+            mod = enums['swizzles_16_bit'].default
+            val = enums['swizzles_16_bit'].bare_values.index(mod)
+            encoded |= (val << src.offset['widen'])
+
+    operands = operands[len(ins.srcs):]
+
+    for i, (op, imm) in enumerate(zip(operands, ins.immediates)):
+        if op[0] == '#':
+            die_if(imm.name != 'constant', "Wrong syntax for immediate")
+            parts = [imm.name, op[1:]]
+        else:
+            parts = op.split(':')
+            die_if(len(parts) != 2, f"Wrong syntax for immediate, wrong number of colons in {op}")
+            die_if(parts[0] != imm.name, f"Wrong immediate, expected {imm.name}, got {parts[0]}")
+
+        if imm.signed:
+            minimum = -(1 << (imm.size - 1))
+            maximum = +(1 << (imm.size - 1)) - 1
+        else:
+            minimum = 0
+            maximum = (1 << imm.size) - 1
+
+        val = parse_int(parts[1], minimum, maximum)
+
+        if val < 0:
+            # Sign extends
+            val = (1 << imm.size) + val
+
+        encoded |= (val << imm.start)
+
+    operands = operands[len(ins.immediates):]
+
+    # Encode the operation itself
+    encoded |= (ins.opcode << 48)
+    encoded |= (ins.opcode2 << ins.secondary_shift)
+
+    # Encode modifiers
+    has_action = False
+    for mod in mods:
+        if len(mod) == 0:
+            continue
+
+        if mod in enums['action'].bare_values:
+            die_if(has_action, "Multiple actions specified")
+            has_action = True
+            encoded |= (enums['action'].bare_values.index(mod) << 59)
+            encoded |= (1 << 62) # Action, not wait
+        elif mod.startswith('wait'):
+            die_if(has_action, "Multiple actions specified")
+            has_action = True
+
+            slots = mod[len('wait'):]
+            try:
+                slots = set([int(x) for x in slots])
+            except ValueError:
+                die(f"Expected slots in {mod}")
+
+            known_slots = set([0, 1, 2])
+            die_if(not slots.issubset(known_slots), f"Unknown slots in {mod}")
+
+            if 0 in slots:
+                encoded |= (1 << 59)
+            if 1 in slots:
+                encoded |= (1 << 60)
+            if 2 in slots:
+                encoded |= (1 << 61)
+        elif mod in enums['immediate_mode'].bare_values:
+            pass # handled specially
+        else:
+            candidates = [c for c in ins.modifiers if mod in c.bare_values]
+
+            die_if(len(candidates) == 0, f"Invalid modifier {mod} used")
+            assert(len(candidates) == 1) # No ambiguous modifiers
+            opts = candidates[0]
+
+            value = opts.bare_values.index(mod)
+            assert(value is not None)
+
+            die_if(opts.name in modifier_map, f"{opts.name} specified twice")
+            modifier_map[opts.name] = value
+
+    for mod in ins.modifiers:
+        value = modifier_map.get(mod.name, mod.default)
+        die_if(value is None, f"Missing required modifier {mod.name}")
+
+        assert(value < (1 << mod.size))
+        encoded |= (value << mod.start)
+
+    encoded |= (enums['immediate_mode'].bare_values.index(immediate_mode) << 57)
+    return encoded
+
+if __name__ == "__main__":
+    # Provide commandline interface
+    parser = argparse.ArgumentParser(description='Assemble Valhall shaders')
+    parser.add_argument('infile', nargs='?', type=argparse.FileType('r'),
+                        default=sys.stdin)
+    parser.add_argument('outfile', type=argparse.FileType('wb'))
+    args = parser.parse_args()
+
+    lines = args.infile.read().strip().split('\n')
+    lines = [l for l in lines if len(l) > 0 and l[0] != '#']
+
+    packed = b''.join([struct.pack('<Q', parse_asm(ln)) for ln in lines])
+    args.outfile.write(packed)



More information about the mesa-commit mailing list