247 lines
8.6 KiB
Python
247 lines
8.6 KiB
Python
# Copyright 2025 The JAX Authors. All Rights Reserved.
|
|
#
|
|
# Licensed under the Apache License, Version 2.0 (the "License");
|
|
# you may not use this file except in compliance with the License.
|
|
# You may obtain a copy of the License at
|
|
#
|
|
# http://www.apache.org/licenses/LICENSE-2.0
|
|
#
|
|
# Unless required by applicable law or agreed to in writing, software
|
|
# distributed under the License is distributed on an "AS IS" BASIS,
|
|
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
|
# See the License for the specific language governing permissions and
|
|
# limitations under the License.
|
|
# ==============================================================================
|
|
|
|
import itertools
|
|
from jax.experimental.mosaic.gpu import fragmented_array as fa
|
|
from jaxlib.mlir import ir
|
|
from jaxlib.mlir.dialects import llvm
|
|
from jaxlib.mlir.dialects import vector
|
|
import numpy as np
|
|
from . import utils
|
|
|
|
|
|
SUPPORTED_F8_TYPES = (ir.Float8E4M3FNType, ir.Float8E5M2Type)
|
|
|
|
|
|
class MMALayouts:
|
|
"""Container for MMA layouts, providing a convenient way to create
|
|
layouts for MMA operands based on warp configuration.
|
|
"""
|
|
|
|
def __init__(self, element_type: ir.Type):
|
|
elems_per_reg = 32 // utils.bitwidth(element_type)
|
|
k = 8 * elems_per_reg
|
|
sub_k = 4 * elems_per_reg
|
|
self.lhs = fa.TiledLayout(
|
|
fa.Tiling(((64, k), (16, sub_k), (8, sub_k), (elems_per_reg,))),
|
|
warp_dims=(-7,),
|
|
lane_dims=(-3, -2),
|
|
vector_dim=-1,
|
|
)
|
|
self.rhs = fa.TiledLayout(
|
|
fa.Tiling(((8, k), (8, sub_k), (elems_per_reg,))),
|
|
warp_dims=(fa.Replicated(4),),
|
|
lane_dims=(-3, -2),
|
|
vector_dim=-1,
|
|
)
|
|
self.acc = fa.TiledLayout(
|
|
fa.Tiling(((64, 8), (16, 8), (8, 8), (2,))),
|
|
warp_dims=(-7,),
|
|
lane_dims=(-3, -2),
|
|
vector_dim=-1,
|
|
)
|
|
|
|
|
|
def _ptx_dtype_str(dtype: ir.Type, *, is_signed: bool | None = None) -> str:
|
|
if isinstance(dtype, ir.Float8E4M3FNType):
|
|
return "e4m3"
|
|
elif isinstance(dtype, ir.Float8E5M2Type):
|
|
return "e5m2"
|
|
elif isinstance(dtype, ir.IntegerType):
|
|
if is_signed is None:
|
|
raise ValueError("is_signed must be specified for integer types")
|
|
prefix = "s" if is_signed else "u"
|
|
return f"{prefix}{dtype.width}"
|
|
return str(dtype)
|
|
|
|
|
|
def _mma_single_tile(
|
|
acc: fa.FragmentedArray, a: fa.FragmentedArray, b: fa.FragmentedArray
|
|
) -> fa.FragmentedArray:
|
|
"""Performs `acc + a @ b.T` using warp level MMA instructions."""
|
|
i32 = ir.IntegerType.get_signless(32)
|
|
|
|
k_tile = 256 // utils.bitwidth(a.mlir_dtype)
|
|
assert a.shape == (64, k_tile)
|
|
assert b.shape == (8, k_tile)
|
|
assert acc.shape == (64, 8)
|
|
assert a.mlir_dtype == b.mlir_dtype
|
|
is_integer = isinstance(a.mlir_dtype, ir.IntegerType)
|
|
assert acc.mlir_dtype == i32 if is_integer else ir.F32Type.get()
|
|
assert acc.is_signed in {None, True}
|
|
assert (
|
|
isinstance(acc.layout, fa.TiledLayout)
|
|
and isinstance(a.layout, fa.TiledLayout)
|
|
and isinstance(b.layout, fa.TiledLayout)
|
|
)
|
|
num_acc_regs, num_a_regs, num_b_regs = 4, 4, 2
|
|
|
|
acc_regs = [
|
|
vector.extract(
|
|
reg,
|
|
dynamic_position=[],
|
|
static_position=ir.DenseI64ArrayAttr.get([pos]),
|
|
)
|
|
for reg in acc.registers.flatten()
|
|
for pos in range(acc.layout.vector_length)
|
|
]
|
|
a_regs = [utils.bitcast(r, i32) for r in a.registers.flatten()]
|
|
b_regs = [utils.bitcast(r, i32) for r in b.registers.flatten()]
|
|
|
|
# Make sure we have the right number of registers for the instruction.
|
|
assert len(a_regs) == 4
|
|
assert len(acc_regs) == 4
|
|
assert len(b_regs) == 2
|
|
|
|
a_ptx_dtype = _ptx_dtype_str(a.mlir_dtype, is_signed=a.is_signed)
|
|
b_ptx_dtype = _ptx_dtype_str(b.mlir_dtype, is_signed=b.is_signed)
|
|
acc_ptx_dtype = "s32" if is_integer else "f32"
|
|
acc_constraint = "r" if is_integer else "f"
|
|
instr = f"mma.sync.aligned.m16n8k{k_tile}.row.col.{acc_ptx_dtype}.{a_ptx_dtype}.{b_ptx_dtype}.{acc_ptx_dtype}"
|
|
counter = itertools.count()
|
|
n_regs_str = lambda n: (
|
|
"{" + ",".join([f"${next(counter)}" for _ in range(n)]) + "}"
|
|
)
|
|
out_regs_str = n_regs_str(num_acc_regs)
|
|
a_regs_str = n_regs_str(num_a_regs)
|
|
b_regs_str = n_regs_str(num_b_regs)
|
|
c_regs_str = n_regs_str(num_acc_regs)
|
|
ptx = f"{instr} {out_regs_str}, {a_regs_str}, {b_regs_str}, {c_regs_str};"
|
|
# See: https://llvm.org/docs/LangRef.html#inline-assembler-expressions
|
|
constraints = (
|
|
f"{','.join([f'={acc_constraint}']*num_acc_regs)},"
|
|
f"{','.join(['r']*num_a_regs)},"
|
|
f"{','.join(['r']*num_b_regs)},"
|
|
f"{','.join([acc_constraint]*num_acc_regs)}"
|
|
)
|
|
|
|
in_operands = [*a_regs, *b_regs, *acc_regs]
|
|
acc_struct_type = ir.Type.parse(
|
|
f"!llvm.struct<({','.join(str(acc.mlir_dtype) for _ in acc_regs)})>"
|
|
)
|
|
out_regs_struct = llvm.inline_asm(
|
|
acc_struct_type,
|
|
in_operands,
|
|
ptx,
|
|
constraints,
|
|
has_side_effects=False,
|
|
)
|
|
assert isinstance(out_regs_struct, ir.Value)
|
|
out_regs = [
|
|
llvm.extractvalue(acc.mlir_dtype, out_regs_struct, [i])
|
|
for i in range(len(acc_regs))
|
|
]
|
|
vec_regs = []
|
|
vec_undef = llvm.mlir_undef(ir.VectorType.get((2,), acc.mlir_dtype))
|
|
for first, second in zip(out_regs[::2], out_regs[1::2]):
|
|
vec = llvm.insertelement(vec_undef, first, position=utils.c(0, i32))
|
|
vec = llvm.insertelement(vec, second, position=utils.c(1, i32))
|
|
vec_regs.append(vec)
|
|
out_regs = np.asarray(vec_regs, dtype=object).reshape(acc.registers.shape)
|
|
return fa.FragmentedArray(
|
|
_registers=out_regs, _layout=acc.layout, _is_signed=acc.is_signed
|
|
)
|
|
|
|
|
|
def mma(
|
|
acc: fa.FragmentedArray,
|
|
a: fa.FragmentedArray,
|
|
b: fa.FragmentedArray,
|
|
) -> fa.FragmentedArray:
|
|
"""Computes `acc + a @ b.T` using synchronouse MMA instructions.
|
|
|
|
All operands must have `TiledLayout`s. The layouts must be generated
|
|
by the `MMALayouts` class, which ensures that the tiles are mapped
|
|
to the warps correctly.
|
|
|
|
Args:
|
|
acc: A `FragmentedArray` with a `TiledLayout` generated from
|
|
`MMALayouts.acc`.
|
|
a: A `FragmentedArray` with a `TiledLayout` generated from
|
|
`MMALayouts.lhs`.
|
|
b: A `FragmentedArray` with a `TiledLayout` generated from `MMALayouts.rhs`.
|
|
|
|
Returns:
|
|
A new `FragmentedArray` with the result of the computation with
|
|
the same type as `acc`.
|
|
"""
|
|
|
|
(m, k) = a.shape
|
|
(n, k2) = b.shape
|
|
(m2, n2) = acc.shape
|
|
|
|
if m != m2:
|
|
raise ValueError(f"M mismatch: {m} != {m2}")
|
|
if n != n2:
|
|
raise ValueError(f"N mismatch: {n} != {n2}")
|
|
if k != k2:
|
|
raise ValueError(f"K mismatch: {k} != {k2}")
|
|
|
|
# todo(cperivol): A tile shape can have dimensions that are higher
|
|
# multiples of the mma op size as long as those dimensions are not
|
|
# sharded across warps.
|
|
i4 = ir.IntegerType.get_signless(4)
|
|
i8 = ir.IntegerType.get_signless(8)
|
|
i32 = ir.IntegerType.get_signless(32)
|
|
bf16 = ir.BF16Type.get()
|
|
f16 = ir.F16Type.get()
|
|
f8e4m3fn = ir.Float8E4M3FNType.get()
|
|
f8e5m2 = ir.Float8E5M2Type.get()
|
|
if (element_type := a.mlir_dtype) != b.mlir_dtype:
|
|
raise ValueError(f"Dtype mismatch: {a.mlir_dtype} != {b.mlir_dtype}")
|
|
if element_type not in (bf16, f16, f8e4m3fn, f8e5m2, i8, i4):
|
|
raise NotImplementedError(f"Unsupported operand type: {element_type}")
|
|
if isinstance(element_type, ir.IntegerType):
|
|
if acc.mlir_dtype != i32:
|
|
raise NotImplementedError("Only s32 accumulator supported for integer operands.")
|
|
if not acc.is_signed:
|
|
raise ValueError("Only signed accumulator supported for integer operands.")
|
|
elif acc.mlir_dtype != ir.F32Type.get():
|
|
raise NotImplementedError("Only f32 accumulator supported for floating operands.")
|
|
|
|
layouts = MMALayouts(element_type)
|
|
if layouts.lhs != a.layout:
|
|
raise ValueError("Expected MMALayouts.lhs layout for A")
|
|
if layouts.rhs != b.layout:
|
|
raise ValueError("Expected MMALayouts.rhs layout for B")
|
|
if layouts.acc != acc.layout:
|
|
raise ValueError("Expected MMALayouts.acc layout for acc")
|
|
|
|
assert isinstance(a.layout, fa.TiledLayout)
|
|
assert isinstance(b.layout, fa.TiledLayout)
|
|
assert isinstance(acc.layout, fa.TiledLayout)
|
|
m_tile, k_tile = a.layout.base_tile_shape
|
|
n_tile, k_tile2 = b.layout.base_tile_shape
|
|
m_tile2, n_tile2 = acc.layout.base_tile_shape
|
|
|
|
assert k_tile == k_tile2
|
|
assert m_tile2 == m_tile
|
|
assert n_tile2 == n_tile
|
|
|
|
num_m_tiles, num_n_tiles, num_k_tiles = m // m_tile, n // n_tile, k // k_tile
|
|
|
|
# Do not modify the accumualtor itself.
|
|
acc = acc.copy()
|
|
s = lambda idx, length: slice(idx * length, (idx + 1) * length)
|
|
for k_idx in range(num_k_tiles):
|
|
for m_idx in range(num_m_tiles):
|
|
for n_idx in range(num_n_tiles):
|
|
ms = s(m_idx, m_tile)
|
|
ns = s(n_idx, n_tile)
|
|
ks = s(k_idx, k_tile)
|
|
acc[ms, ns] = _mma_single_tile(acc[ms, ns], a[ms, ks], b[ns, ks])
|
|
|
|
return acc
|