submission 779785

#!POPCORN leaderboard matmul_v2
#!POPCORN gpu H100

from task import input_t, output_t
import triton
import triton.language as tl


@triton.autotune(
    configs=[
        triton.Config(
            {"BLOCK_M": 128, "BLOCK_N": 128, "BLOCK_K": 32, "GROUP_SIZE_M": 8},
            num_stages=4, num_warps=8,
        ),
        triton.Config(
            {"BLOCK_M": 128, "BLOCK_N": 256, "BLOCK_K": 64, "GROUP_SIZE_M": 8},
            num_stages=3, num_warps=8,
        ),
        triton.Config(
            {"BLOCK_M": 64, "BLOCK_N": 128, "BLOCK_K": 32, "GROUP_SIZE_M": 8},
            num_stages=4, num_warps=4,
        ),
        triton.Config(
            {"BLOCK_M": 128, "BLOCK_N": 64, "BLOCK_K": 32, "GROUP_SIZE_M": 8},
            num_stages=4, num_warps=4,
        ),
    ],
    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_M: tl.constexpr,
    BLOCK_N: tl.constexpr,
    BLOCK_K: tl.constexpr,
    GROUP_SIZE_M: tl.constexpr,
):
    # L2-cache-friendly (pid_m, pid_n) swizzle, standard Triton tutorial pattern.
    pid = tl.program_id(0)
    num_pid_m = tl.cdiv(M, BLOCK_M)
    num_pid_n = tl.cdiv(N, BLOCK_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 % num_pid_in_group) % group_size_m)
    pid_n = (pid % num_pid_in_group) // group_size_m

    # Tile row/col offsets.
    offs_am = (pid_m * BLOCK_M + tl.arange(0, BLOCK_M)) % M
    offs_bn = (pid_n * BLOCK_N + tl.arange(0, BLOCK_N)) % N
    offs_k = tl.arange(0, BLOCK_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)

    acc = tl.zeros((BLOCK_M, BLOCK_N), dtype=tl.float32)
    for k in range(0, tl.cdiv(K, BLOCK_K)):
        k_mask = offs_k[None, :] < K - k * BLOCK_K
        a = tl.load(a_ptrs, mask=k_mask, other=0.0)
        k_mask_b = offs_k[:, None] < K - k * BLOCK_K
        b = tl.load(b_ptrs, mask=k_mask_b, other=0.0)
        acc = tl.dot(a, b, acc=acc, input_precision="ieee")
        a_ptrs += BLOCK_K * stride_ak
        b_ptrs += BLOCK_K * stride_bk

    # Write back to pre-allocated output, masking M/N edges.
    offs_cm = pid_m * BLOCK_M + tl.arange(0, BLOCK_M)
    offs_cn = pid_n * BLOCK_N + tl.arange(0, BLOCK_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, acc.to(c_ptr.dtype.element_ty), mask=c_mask)


def custom_kernel(data: input_t) -> output_t:
    A, B, output = data
    M, K = A.shape
    K2, N = B.shape
    grid = lambda META: (
        triton.cdiv(M, META["BLOCK_M"]) * triton.cdiv(N, META["BLOCK_N"]),
    )
    matmul_kernel[grid](
        A, B, output,
        M, N, K,
        A.stride(0), A.stride(1),
        B.stride(0), B.stride(1),
        output.stride(0), output.stride(1),
    )
    return output