triton_kernel.py

# (c) Meta Platforms, Inc. and affiliates. Confidential and proprietary.

import functools

import torch

import triton  # @manual
import triton.language as tl  # @manual

torch.set_float32_matmul_precision("high")


@triton.autotune(
    configs=[
        triton.Config(
            {
                "BLOCK_SIZE_M": 128,
                "BLOCK_SIZE_N": 256,
                "BLOCK_SIZE_K": 16,
                "GROUP_SIZE_M": 8,
            },
            num_stages=3,
            num_warps=8,
        ),
        triton.Config(
            {
                "BLOCK_SIZE_M": 64,
                "BLOCK_SIZE_N": 256,
                "BLOCK_SIZE_K": 16,
                "GROUP_SIZE_M": 8,
            },
            num_stages=4,
            num_warps=4,
        ),
        triton.Config(
            {
                "BLOCK_SIZE_M": 128,
                "BLOCK_SIZE_N": 128,
                "BLOCK_SIZE_K": 16,
                "GROUP_SIZE_M": 8,
            },
            num_stages=4,
            num_warps=4,
        ),
        triton.Config(
            {
                "BLOCK_SIZE_M": 128,
                "BLOCK_SIZE_N": 64,
                "BLOCK_SIZE_K": 16,
                "GROUP_SIZE_M": 8,
            },
            num_stages=4,
            num_warps=4,
        ),
        triton.Config(
            {
                "BLOCK_SIZE_M": 64,
                "BLOCK_SIZE_N": 128,
                "BLOCK_SIZE_K": 16,
                "GROUP_SIZE_M": 8,
            },
            num_stages=4,
            num_warps=4,
        ),
        triton.Config(
            {
                "BLOCK_SIZE_M": 128,
                "BLOCK_SIZE_N": 32,
                "BLOCK_SIZE_K": 16,
                "GROUP_SIZE_M": 8,
            },
            num_stages=4,
            num_warps=4,
        ),
        triton.Config(
            {
                "BLOCK_SIZE_M": 64,
                "BLOCK_SIZE_N": 32,
                "BLOCK_SIZE_K": 16,
                "GROUP_SIZE_M": 8,
            },
            num_stages=5,
            num_warps=2,
        ),
        triton.Config(
            {
                "BLOCK_SIZE_M": 32,
                "BLOCK_SIZE_N": 64,
                "BLOCK_SIZE_K": 16,
                "GROUP_SIZE_M": 8,
            },
            num_stages=5,
            num_warps=2,
        ),
    ],
    key=["M", "N", "K"],
)
@triton.jit
def matmul_kernel(
    a_ptr,
    b_ptr,
    c_ptr,
    M,
    N,
    K,
    stride_am,
    stride_ak,  #
    stride_bk,
    stride_bn,  #
    stride_cm,
    stride_cn,
    BLOCK_SIZE_M: tl.constexpr,
    BLOCK_SIZE_N: tl.constexpr,
    BLOCK_SIZE_K: tl.constexpr,  #
    GROUP_SIZE_M: tl.constexpr,  #
):
    pid = tl.program_id(axis=0)
    num_pid_m = tl.cdiv(M, BLOCK_SIZE_M)
    num_pid_n = tl.cdiv(N, BLOCK_SIZE_N)
    num_pid_in_group = GROUP_SIZE_M * num_pid_n
    group_id = pid // num_pid_in_group
    first_pid_m = group_id * GROUP_SIZE_M
    group_size_m = min(num_pid_m - first_pid_m, GROUP_SIZE_M)
    pid_m = first_pid_m + (pid % group_size_m)
    pid_n = (pid % num_pid_in_group) // group_size_m

    offs_am = (pid_m * BLOCK_SIZE_M + tl.arange(0, BLOCK_SIZE_M)) % M
    offs_bn = (pid_n * BLOCK_SIZE_N + tl.arange(0, BLOCK_SIZE_N)) % N
    offs_k = tl.arange(0, BLOCK_SIZE_K)
    a_ptrs = a_ptr + (offs_am[:, None] * stride_am + offs_k[None, :] * stride_ak)
    b_ptrs = b_ptr + (offs_k[:, None] * stride_bk + offs_bn[None, :] * stride_bn)

    accumulator = tl.zeros((BLOCK_SIZE_M, BLOCK_SIZE_N), dtype=tl.float32)
    for k in range(0, tl.cdiv(K, BLOCK_SIZE_K)):
        a = tl.load(a_ptrs, mask=offs_k[None, :] < K - k * BLOCK_SIZE_K, other=0.0)
        b = tl.load(b_ptrs, mask=offs_k[:, None] < K - k * BLOCK_SIZE_K, other=0.0)
        accumulator = tl.dot(a, b, accumulator)
        a_ptrs += BLOCK_SIZE_K * stride_ak
        b_ptrs += BLOCK_SIZE_K * stride_bk
    c = accumulator.to(c_ptr.dtype.element_ty)

    offs_cm = pid_m * BLOCK_SIZE_M + tl.arange(0, BLOCK_SIZE_M)
    offs_cn = pid_n * BLOCK_SIZE_N + tl.arange(0, BLOCK_SIZE_N)
    c_ptrs = c_ptr + stride_cm * offs_cm[:, None] + stride_cn * offs_cn[None, :]
    c_mask = (offs_cm[:, None] < M) & (offs_cn[None, :] < N)
    tl.store(c_ptrs, c, mask=c_mask)


@triton.jit
def leaky_relu(x):
    x = x + 1
    return tl.where(x >= 0, x, 0.01 * x)


@functools.lru_cache()
def print_once(s):
    print(s)


precompiled_kernel = None
try:
    precompiled_kernel = triton.compiler.compile(
        "matmul_kernel.ttgir"
    )
except Exception:
    pass


def matmul(a, b, precompiled=False):
    # Check constraints.
    assert a.shape[1] == b.shape[0], "Incompatible dimensions"
    assert a.is_contiguous(), "Matrix A must be contiguous"
    M, K = a.shape
    K, N = b.shape
    # Allocates output.
    c = torch.empty((M, N), device=a.device, dtype=torch.float32)
    # 1D launch kernel where each block gets its own program.
    if precompiled and precompiled_kernel:
        precompiled_kernel[((M // 128) * (N // 256)), 1, 1](
            a, b, c, M, N, K, a.stride(0), b.stride(0), c.stride(0)
        )
    else:

        def grid(META):
            return (
                triton.cdiv(M, META["BLOCK_SIZE_M"])
                * triton.cdiv(N, META["BLOCK_SIZE_N"]),
            )

        matmul_kernel[grid](
            a,
            b,
            c,  #
            M,
            N,
            K,  #
            a.stride(0),
            a.stride(1),  #
            b.stride(0),
            b.stride(1),  #
            c.stride(0),
            c.stride(1),  #
        )
    # print_once(bin.asm["ttgir"])
    return c