submission 226087

#!POPCORN leaderboard nvfp4_dual_gemm
#!POPCORN gpu NVIDIA

import torch
from task import input_t, output_t
from torch.utils.cpp_extension import load_inline

# Shared CUDA source with both kernels.
cuda_src = """
#include <cudaTypedefs.h>
#include <cuda_fp16.h>
#include <cuda_fp8.h>

#include <torch/library.h>
#include <ATen/core/Tensor.h>

constexpr int WARP_SIZE = 32;
constexpr int MMA_K = 64;

constexpr uint64_t EVICT_FIRST = 0x12F0000000000000;
constexpr uint64_t EVICT_LAST = 0x14F0000000000000;

__device__ inline
constexpr uint64_t desc_encode(uint64_t x) { return (x & 0x3'FFFFULL) >> 4ULL; };

__device__ inline uint32_t elect_sync() {
    uint32_t pred = 0;
    asm volatile(
        "{\\n\\t"
        ".reg .pred %%px;\\n\\t"
        "elect.sync _|%%px, %1;\\n\\t"
        "@%%px mov.s32 %0, 1;\\n\\t"
        "}"
        : "+r"(pred)
        : "r"(0xFFFFFFFF)
    );
    return pred;
}

__device__ inline
void mbarrier_init(int mbar_addr, int count) {
    asm volatile("mbarrier.init.shared::cta.b64 [%0], %1;" :: "r"(mbar_addr), "r"(count));
}

__device__
void mbarrier_wait(int mbar_addr, int phase) {
    uint32_t ticks = 0x989680;
    asm volatile(
        "{\\n\\t"
        ".reg .pred P1;\\n\\t"
        "LAB_WAIT:\\n\\t"
        "mbarrier.try_wait.parity.acquire.cta.shared::cta.b64 P1, [%0], %1, %2;\\n\\t"
        "@P1 bra.uni DONE;\\n\\t"
        "bra.uni LAB_WAIT;\\n\\t"
        "DONE:\\n\\t"
        "}"
        :: "r"(mbar_addr), "r"(phase), "r"(ticks)
    );
}

__device__ inline
void mbarrier_arrive(int mbar_addr) {
    asm volatile("mbarrier.arrive.release.cta.shared::cta.b64 _, [%0];" :: "r"(mbar_addr) : "memory");
}

__device__ inline
void mbarrier_expect_tx(int mbar_addr, int size) {
    asm volatile("mbarrier.arrive.expect_tx.release.cta.shared::cta.b64 _, [%0], %1;"
                        :: "r"(mbar_addr), "r"(size) : "memory");
}

__device__ inline
void tma_copy(int dst, const void *src, int size, int mbar_addr, uint64_t cache_policy) {
    asm volatile("cp.async.bulk.shared::cta.global.mbarrier::complete_tx::bytes.L2::cache_hint [%0], [%1], %2, [%3], %4;"
                        :: "r"(dst), "l"(src), "r"(size), "r"(mbar_addr), "l"(cache_policy));
}

__device__ inline
void tma_load_3d(int dst, const void *tmap_ptr, int x, int y, int z, int mbar_addr, uint64_t cache_policy) {
    asm volatile("cp.async.bulk.tensor.3d.shared::cta.global.mbarrier::complete_tx::bytes.cta_group::1.L2::cache_hint "
                        "[%0], [%1, {%2, %3, %4}], [%5], %6;"
                        :: "r"(dst), "l"(tmap_ptr), "r"(x), "r"(y), "r"(z), "r"(mbar_addr), "l"(cache_policy)
                        : "memory");
}

template <int BLOCK_M, int BLOCK_N, int BLOCK_K>
__device__ inline
void issue_tma_interleaved(
    int smem, int stage_id, int iter_k,
    const CUtensorMap *A_tmap, const CUtensorMap *B1_tmap, const CUtensorMap *B2_tmap,
    const char *SFA_ptr, const char *SFB1_ptr, const char *SFB2_ptr,
    int off_m, int off_n, int K,
    int mbar_addr, uint64_t cache_A, uint64_t cache_B
) {
    constexpr int A_size = BLOCK_M * BLOCK_K / 2;
    constexpr int B_size = BLOCK_N * BLOCK_K / 2;
    constexpr int SF_size = 128 * BLOCK_K / 16;
    constexpr int STAGE_SIZE = A_size + B_size * 2 + SF_size * 3;

    const int A_smem = smem + stage_id * STAGE_SIZE;
    const int B1_smem = A_smem + A_size;
    const int B2_smem = B1_smem + B_size;
    const int SFA_smem = B2_smem + B_size;
    const int SFB1_smem = SFA_smem + SF_size;
    const int SFB2_smem = SFB1_smem + SF_size;

    const int off_k = iter_k * BLOCK_K;

    tma_load_3d(A_smem, A_tmap, 0, off_m, off_k / 256, mbar_addr, cache_A);
    tma_load_3d(B1_smem, B1_tmap, 0, off_n, off_k / 256, mbar_addr, cache_B);
    tma_load_3d(B2_smem, B2_tmap, 0, off_n, off_k / 256, mbar_addr, cache_B);

    const int rest_k = K / 16 / 4;
    const char *SFA_src = SFA_ptr + ((off_m / 128) * rest_k + off_k / (16 * 4)) * 512;
    const char *SFB1_src = SFB1_ptr + ((off_n / 128) * rest_k + off_k / (16 * 4)) * 512;
    const char *SFB2_src = SFB2_ptr + ((off_n / 128) * rest_k + off_k / (16 * 4)) * 512;

    tma_copy(SFA_smem, SFA_src, SF_size, mbar_addr, cache_A);
    tma_copy(SFB1_smem, SFB1_src, SF_size, mbar_addr, cache_B);
    tma_copy(SFB2_smem, SFB2_src, SF_size, mbar_addr, cache_B);

    mbarrier_expect_tx(mbar_addr, STAGE_SIZE);
}

__device__ inline
void tcgen05_cp_nvfp4(int taddr, uint64_t s_desc) {
    asm volatile("tcgen05.cp.cta_group::1.32x128b.warpx4 [%0], %1;" :: "r"(taddr), "l"(s_desc));
}

__device__ inline
void tcgen05_mma_nvfp4(uint64_t a_desc, uint64_t b_desc, uint32_t i_desc,
    int scale_A_tmem, int scale_B_tmem, int enable_input_d, int d_tmem) {
    asm volatile(
        "{\\n\\t"
        ".reg .pred p;\\n\\t"
        "setp.ne.b32 p, %6, 0;\\n\\t"
        "tcgen05.mma.cta_group::1.kind::mxf4nvf4.block_scale.block16 [%0], %1, %2, %3, [%4], [%5], p;\\n\\t"
        "}"
        :: "r"(d_tmem), "l"(a_desc), "l"(b_desc), "r"(i_desc),
             "r"(scale_A_tmem), "r"(scale_B_tmem), "r"(enable_input_d)
    );
}

__device__ inline
void tcgen05_commit(int mbar_addr) {
    asm volatile("tcgen05.commit.cta_group::1.mbarrier::arrive::one.shared::cluster.b64 [%0];"
                        :: "r"(mbar_addr) : "memory");
}

struct SHAPE {
    static constexpr char _16x256b[] = ".16x256b";
};

struct NUM {
    static constexpr char x8[]  = ".x8";
    static constexpr char x16[] = ".x16";
};

template <const char *SHAPE, const char *NUM>
__device__ inline
void tcgen05_ld_32regs(float *tmp, int row, int col) {
    asm volatile("tcgen05.ld.sync.aligned%33%34.b32 "
        "{ %0,  %1,  %2,  %3,  %4,  %5,  %6,  %7, "
        "  %8,  %9, %10, %11, %12, %13, %14, %15, "
        " %16, %17, %18, %19, %20, %21, %22, %23, "
        " %24, %25, %26, %27, %28, %29, %30, %31}, [%32];"
        : "=f"(tmp[ 0]), "=f"(tmp[ 1]), "=f"(tmp[ 2]), "=f"(tmp[ 3]), "=f"(tmp[ 4]), "=f"(tmp[ 5]), "=f"(tmp[ 6]), "=f"(tmp[ 7]),
            "=f"(tmp[ 8]), "=f"(tmp[ 9]), "=f"(tmp[10]), "=f"(tmp[11]), "=f"(tmp[12]), "=f"(tmp[13]), "=f"(tmp[14]), "=f"(tmp[15]),
            "=f"(tmp[16]), "=f"(tmp[17]), "=f"(tmp[18]), "=f"(tmp[19]), "=f"(tmp[20]), "=f"(tmp[21]), "=f"(tmp[22]), "=f"(tmp[23]),
            "=f"(tmp[24]), "=f"(tmp[25]), "=f"(tmp[26]), "=f"(tmp[27]), "=f"(tmp[28]), "=f"(tmp[29]), "=f"(tmp[30]), "=f"(tmp[31])
        : "r"((row << 16) | col), "C"(SHAPE), "C"(NUM));
}

template <const char *SHAPE, const char *NUM>
__device__ inline
void tcgen05_ld_64regs(float *tmp, int row, int col) {
    asm volatile("tcgen05.ld.sync.aligned%65%66.b32 "
                "{ %0,  %1,  %2,  %3,  %4,  %5,  %6,  %7, "
                "  %8,  %9, %10, %11, %12, %13, %14, %15, "
                " %16, %17, %18, %19, %20, %21, %22, %23, "
                " %24, %25, %26, %27, %28, %29, %30, %31, "
                " %32, %33, %34, %35, %36, %37, %38, %39, "
                " %40, %41, %42, %43, %44, %45, %46, %47, "
                " %48, %49, %50, %51, %52, %53, %54, %55, "
                " %56, %57, %58, %59, %60, %61, %62, %63}, [%64];"
                : "=f"(tmp[ 0]), "=f"(tmp[ 1]), "=f"(tmp[ 2]), "=f"(tmp[ 3]), "=f"(tmp[ 4]), "=f"(tmp[ 5]), "=f"(tmp[ 6]), "=f"(tmp[ 7]),
                    "=f"(tmp[ 8]), "=f"(tmp[ 9]), "=f"(tmp[10]), "=f"(tmp[11]), "=f"(tmp[12]), "=f"(tmp[13]), "=f"(tmp[14]), "=f"(tmp[15]),
                    "=f"(tmp[16]), "=f"(tmp[17]), "=f"(tmp[18]), "=f"(tmp[19]), "=f"(tmp[20]), "=f"(tmp[21]), "=f"(tmp[22]), "=f"(tmp[23]),
                    "=f"(tmp[24]), "=f"(tmp[25]), "=f"(tmp[26]), "=f"(tmp[27]), "=f"(tmp[28]), "=f"(tmp[29]), "=f"(tmp[30]), "=f"(tmp[31]),
                    "=f"(tmp[32]), "=f"(tmp[33]), "=f"(tmp[34]), "=f"(tmp[35]), "=f"(tmp[36]), "=f"(tmp[37]), "=f"(tmp[38]), "=f"(tmp[39]),
                    "=f"(tmp[40]), "=f"(tmp[41]), "=f"(tmp[42]), "=f"(tmp[43]), "=f"(tmp[44]), "=f"(tmp[45]), "=f"(tmp[46]), "=f"(tmp[47]),
                    "=f"(tmp[48]), "=f"(tmp[49]), "=f"(tmp[50]), "=f"(tmp[51]), "=f"(tmp[52]), "=f"(tmp[53]), "=f"(tmp[54]), "=f"(tmp[55]),
                    "=f"(tmp[56]), "=f"(tmp[57]), "=f"(tmp[58]), "=f"(tmp[59]), "=f"(tmp[60]), "=f"(tmp[61]), "=f"(tmp[62]), "=f"(tmp[63])
                : "r"((row << 16) | col), "C"(SHAPE), "C"(NUM));
}

// Convenient wrappers for epilogue loads
__device__ inline void tcgen05_ld_16x256bx8(float *tmp, int row, int col) { tcgen05_ld_32regs<SHAPE::_16x256b, NUM::x8>(tmp, row, col); }
__device__ inline void tcgen05_ld_16x256bx16(float *tmp, int row, int col) { tcgen05_ld_64regs<SHAPE::_16x256b, NUM::x16>(tmp, row, col); }

__device__ inline
float silu(float x) {
    //return x / (1.0f + expf(-x));
    //return __fdividef(x, 1.0f + __expf(-x));
    return x * __fdividef(1.0f, (1.0f + expf(-x)));

}

// ============================================================================
// EPILOGUE STRATEGY 1: Baseline (current implementation)
// Uses .16x256b, 2 loads per warp slab, 32/64 regs per array
// ============================================================================
template <int BLOCK_N>
__device__ inline
void epilogue_v1_baseline(
    int warp_id, int lane_id,
    int off_m, int off_n,
    int gemm1_tmem, int gemm2_tmem,
    half *C_ptr, int N,
    int done_mbar_addr
) {
    mbarrier_wait(done_mbar_addr, 0);
    asm volatile("tcgen05.fence::after_thread_sync;");
    #pragma unroll
    for (int m = 0; m < 32 / 16; m++) {
        float g1[BLOCK_N / 2];
        float g2[BLOCK_N / 2];

        if constexpr (BLOCK_N == 128) {
            tcgen05_ld_16x256bx16(g1, warp_id * 32 + m * 16, gemm1_tmem);
            tcgen05_ld_16x256bx16(g2, warp_id * 32 + m * 16, gemm2_tmem);
        } else {
            tcgen05_ld_16x256bx8(g1, warp_id * 32 + m * 16, gemm1_tmem);
            tcgen05_ld_16x256bx8(g2, warp_id * 32 + m * 16, gemm2_tmem);
        }
        asm volatile("tcgen05.wait::ld.sync.aligned;");

        #pragma unroll
        for (int i = 0; i < BLOCK_N / 8; i++) {
            const int row0 = warp_id * 32 + m * 16 + lane_id / 4;
            const int row1 = row0 + 8;
            const int col0 = i * 8 + (lane_id % 4) * 2;

            const float s00 = silu(g1[i * 4 + 0]);
            const float s01 = silu(g1[i * 4 + 1]);
            const float s10 = silu(g1[i * 4 + 2]);
            const float s11 = silu(g1[i * 4 + 3]);

            const float v00 = g2[i * 4 + 0] * s00;
            const float v01 = g2[i * 4 + 1] * s01;
            const float v10 = g2[i * 4 + 2] * s10;
            const float v11 = g2[i * 4 + 3] * s11;

            const int out_row0 = off_m + row0;
            const int out_row1 = off_m + row1;
            const int out_col = off_n + col0;

            reinterpret_cast<half2 *>(C_ptr + out_row0 * N + out_col)[0] = __float22half2_rn({v00, v01});
            reinterpret_cast<half2 *>(C_ptr + out_row1 * N + out_col)[0] = __float22half2_rn({v10, v11});
        }
    }

    if (warp_id == 0)
        asm volatile("tcgen05.dealloc.cta_group::1.sync.aligned.b32 %0, %1;" :: "r"(0), "r"(BLOCK_N * 4));
}

void check_cu(CUresult err) {
    if (err == CUDA_SUCCESS) return;
    const char *error_msg_ptr;
    if (cuGetErrorString(err, &error_msg_ptr) != CUDA_SUCCESS)
        error_msg_ptr = "unable to get error string";
    TORCH_CHECK(false, "cuTensorMapEncodeTiled error: ", error_msg_ptr);
}

void init_AB_tmap(
    CUtensorMap *tmap,
    const char *ptr,
    uint64_t global_height, uint64_t global_width,
    uint32_t shared_height, uint32_t shared_width
) {
    constexpr uint32_t rank = 3;
    uint64_t globalDim[rank]       = {256, global_height, global_width / 256};
    uint64_t globalStrides[rank-1] = {global_width / 2, 128};
    uint32_t boxDim[rank]          = {256, shared_height, shared_width / 256};
    uint32_t elementStrides[rank]  = {1, 1, 1};

    auto err = cuTensorMapEncodeTiled(
        tmap,
        CUtensorMapDataType::CU_TENSOR_MAP_DATA_TYPE_16U4_ALIGN8B,
        rank,
        (void *)ptr,
        globalDim,
        globalStrides,
        boxDim,
        elementStrides,
        CUtensorMapInterleave::CU_TENSOR_MAP_INTERLEAVE_NONE,
        CUtensorMapSwizzle::CU_TENSOR_MAP_SWIZZLE_128B,
        CUtensorMapL2promotion::CU_TENSOR_MAP_L2_PROMOTION_NONE,
        CUtensorMapFloatOOBfill::CU_TENSOR_MAP_FLOAT_OOB_FILL_NONE
    );
    check_cu(err);
}

// Interleaved dual GEMM kernel
template <int BLOCK_M, int BLOCK_N, int BLOCK_K, int NUM_STAGES>
__global__
__launch_bounds__(BLOCK_M + 2 * WARP_SIZE)
void dual_gemm_silu_kernel(
    const __grid_constant__ CUtensorMap A_tmap,
    const __grid_constant__ CUtensorMap B1_tmap,
    const __grid_constant__ CUtensorMap B2_tmap,
    const char* __restrict__ SFA_ptr,
    const char* __restrict__ SFB1_ptr,
    const char* __restrict__ SFB2_ptr,
    half* __restrict__ C_ptr,
    int M, int N, int K
) {
    const int tid = threadIdx.x;
    const int bid = blockIdx.x;
    const int lane_id = tid % WARP_SIZE;
    const int warp_id = tid / WARP_SIZE;

    const int grid_n = N / BLOCK_N;
    const int bid_m = bid / grid_n;
    const int bid_n = bid % grid_n;
    const int off_m = bid_m * BLOCK_M;
    const int off_n = bid_n * BLOCK_N;

    constexpr int NUM_WARPS = BLOCK_M / WARP_SIZE + 2;
    const int num_iters = K / BLOCK_K;

    extern __shared__ __align__(1024) char smem_ptr[];
    const int smem = static_cast<int>(__cvta_generic_to_shared(smem_ptr));

    constexpr int A_size = BLOCK_M * BLOCK_K / 2;
    constexpr int B_size = BLOCK_N * BLOCK_K / 2;
    constexpr int SF_size = 128 * BLOCK_K / 16;
    constexpr int STAGE_SIZE = A_size + B_size * 2 + SF_size * 3;

    #pragma nv_diag_suppress static_var_with_dynamic_init
    __shared__ int64_t mbars[NUM_STAGES * 2 + 1];
    const int tma_mbar_addr = static_cast<int>(__cvta_generic_to_shared(mbars));
    const int mma_mbar_addr = tma_mbar_addr + NUM_STAGES * 8;
    const int done_mbar_addr = mma_mbar_addr + NUM_STAGES * 8;

    constexpr int GEMM1_D_TMEM = 0;
    constexpr int GEMM2_D_TMEM = BLOCK_N;
    constexpr int SFA_tmem = BLOCK_N * 2;
    constexpr int SFB1_tmem = SFA_tmem + 4 * (BLOCK_K / MMA_K);
    constexpr int SFB2_tmem = SFB1_tmem + 4 * (BLOCK_K / MMA_K);

    if (warp_id == 0 && elect_sync()) {
        #pragma unroll
        for (int i = 0; i < NUM_STAGES * 2 + 1; i++)
            mbarrier_init(tma_mbar_addr + i * 8, 1);
        asm volatile("fence.mbarrier_init.release.cluster;");
    }
    else if (warp_id == 1) {
        asm volatile("tcgen05.alloc.cta_group::1.sync.aligned.shared::cta.b32 [%0], %1;"
            :: "r"(smem), "r"(BLOCK_N * 4));
    }
    __syncthreads();

    uint64_t cache_A = (M > N) ? EVICT_FIRST : EVICT_LAST;
    uint64_t cache_B = (M > N) ? EVICT_LAST : EVICT_FIRST;

    auto make_desc_AB = [](int addr) -> uint64_t {
        const int SBO = 8 * 128;
        return desc_encode(addr) | (desc_encode(SBO) << 32ULL) | (1ULL << 46ULL) | (2ULL << 61ULL);
    };
    auto make_desc_SF = [](int addr) -> uint64_t {
        const int SBO = 8 * 16;
        return desc_encode(addr) | (desc_encode(SBO) << 32ULL) | (1ULL << 46ULL);
    };

    constexpr uint32_t i_desc = (1U << 7U) | (1U << 10U) | ((uint32_t)BLOCK_N >> 3U << 17U) | ((uint32_t)128 >> 7U << 27U);

    if (warp_id == NUM_WARPS - 2 && elect_sync()) {
        // TMA warp
        #pragma unroll
        for (int iter_k = 0; iter_k < NUM_STAGES && iter_k < num_iters; iter_k++) {
            issue_tma_interleaved<BLOCK_M, BLOCK_N, BLOCK_K>(
                smem, iter_k, iter_k,
                &A_tmap, &B1_tmap, &B2_tmap,
                SFA_ptr, SFB1_ptr, SFB2_ptr,
                off_m, off_n, K, tma_mbar_addr + iter_k * 8, cache_A, cache_B);
        }
        #pragma unroll
        for (int iter_k = NUM_STAGES; iter_k < num_iters; iter_k++) {
            const int stage_id = iter_k % NUM_STAGES;
            mbarrier_wait(mma_mbar_addr + stage_id * 8, (iter_k / NUM_STAGES - 1) % 2);
            issue_tma_interleaved<BLOCK_M, BLOCK_N, BLOCK_K>(
                smem, stage_id, iter_k,
                &A_tmap, &B1_tmap, &B2_tmap,
                SFA_ptr, SFB1_ptr, SFB2_ptr,
                off_m, off_n, K, tma_mbar_addr + stage_id * 8, cache_A, cache_B);
        }
    }
    else if (warp_id == NUM_WARPS - 1 && elect_sync()) {
        // MMA warp
        #pragma unroll
        for (int iter_k = 0; iter_k < num_iters; iter_k++) {
            const int stage_id = iter_k % NUM_STAGES;
            mbarrier_wait(tma_mbar_addr + stage_id * 8, (iter_k / NUM_STAGES) % 2);

            const int A_smem = smem + stage_id * STAGE_SIZE;
            const int B1_smem = A_smem + A_size;
            const int B2_smem = B1_smem + B_size;
            const int SFA_smem = B2_smem + B_size;
            const int SFB1_smem = SFA_smem + SF_size;
            const int SFB2_smem = SFB1_smem + SF_size;

            constexpr uint64_t SF_desc_base = make_desc_SF(0);
            const uint64_t SFA_desc = SF_desc_base + ((uint64_t)SFA_smem >> 4ULL);
            const uint64_t SFB1_desc = SF_desc_base + ((uint64_t)SFB1_smem >> 4ULL);
            const uint64_t SFB2_desc = SF_desc_base + ((uint64_t)SFB2_smem >> 4ULL);

            #pragma unroll
            for (int k = 0; k < BLOCK_K / MMA_K; k++) {
                tcgen05_cp_nvfp4(SFA_tmem + k * 4, SFA_desc + (uint64_t)k * (512ULL >> 4ULL));
                tcgen05_cp_nvfp4(SFB1_tmem + k * 4, SFB1_desc + (uint64_t)k * (512ULL >> 4ULL));
                tcgen05_cp_nvfp4(SFB2_tmem + k * 4, SFB2_desc + (uint64_t)k * (512ULL >> 4ULL));
            }

            // Interleaved GEMM1 and GEMM2 MMA
            #pragma unroll
            for (int k1 = 0; k1 < BLOCK_K / 256; k1++) {
                #pragma unroll
                for (int k2 = 0; k2 < 256 / MMA_K; k2++) {
                    uint64_t a_desc = make_desc_AB(A_smem + k1 * BLOCK_M * 128 + k2 * 32);
                    uint64_t b1_desc = make_desc_AB(B1_smem + k1 * BLOCK_N * 128 + k2 * 32);
                    uint64_t b2_desc = make_desc_AB(B2_smem + k1 * BLOCK_N * 128 + k2 * 32);
                    int k_sf = k1 * 4 + k2;
                    const int scale_A_tmem = SFA_tmem + k_sf * 4 + (bid_m % (128 / BLOCK_M)) * (BLOCK_M / 32);
                    const int scale_B1_tmem = SFB1_tmem + k_sf * 4 + (bid_n % (128 / BLOCK_N)) * (BLOCK_N / 32);
                    const int scale_B2_tmem = SFB2_tmem + k_sf * 4 + (bid_n % (128 / BLOCK_N)) * (BLOCK_N / 32);
                    const int enable_input_d = (k1 == 0 && k2 == 0) ? iter_k : 1;
                    tcgen05_mma_nvfp4(a_desc, b1_desc, i_desc, scale_A_tmem, scale_B1_tmem, enable_input_d, GEMM1_D_TMEM);
                    tcgen05_mma_nvfp4(a_desc, b2_desc, i_desc, scale_A_tmem, scale_B2_tmem, enable_input_d, GEMM2_D_TMEM);
                }
            }

            tcgen05_commit(mma_mbar_addr + stage_id * 8);
        }
        tcgen05_commit(done_mbar_addr);
    }
    else if (tid < BLOCK_M) {
        epilogue_v1_baseline<BLOCK_N>(
            warp_id, lane_id,
            off_m, off_n,
            GEMM1_D_TMEM, GEMM2_D_TMEM,
            C_ptr, N,
            done_mbar_addr
        );
    }
}

// BLOCK_N=64 kernel for m=256 cases (5 stages)
at::Tensor dual_gemm_silu_n64(
    const at::Tensor& A,
    const at::Tensor& B1,
    const at::Tensor& B2,
    const at::Tensor& SFA,
    const at::Tensor& SFB1,
    const at::Tensor& SFB2,
                at::Tensor& C
) {
    const int M = A.size(0);
    const int N = B1.size(0);
    const int K = A.size(1) * 2;

    constexpr int BLOCK_M = 128;
    constexpr int BLOCK_N = 64;
    constexpr int BLOCK_K = 256;
    constexpr int NUM_STAGES = 5;  // 5 stages × 38KB = 190KB (need room for static mbars)

    auto A_ptr   = reinterpret_cast<const char *>(A.data_ptr());
    auto B1_ptr  = reinterpret_cast<const char *>(B1.data_ptr());
    auto B2_ptr  = reinterpret_cast<const char *>(B2.data_ptr());
    auto SFA_ptr = reinterpret_cast<const char *>(SFA.data_ptr());
    auto SFB1_ptr = reinterpret_cast<const char *>(SFB1.data_ptr());
    auto SFB2_ptr = reinterpret_cast<const char *>(SFB2.data_ptr());
    auto C_ptr   = reinterpret_cast<half *>(C.data_ptr());

    CUtensorMap A_tmap, B1_tmap, B2_tmap;
    init_AB_tmap(&A_tmap, A_ptr, M, K, BLOCK_M, BLOCK_K);
    init_AB_tmap(&B1_tmap, B1_ptr, N, K, BLOCK_N, BLOCK_K);
    init_AB_tmap(&B2_tmap, B2_ptr, N, K, BLOCK_N, BLOCK_K);

    int grid = (M / BLOCK_M) * (N / BLOCK_N);
    int tb_size = BLOCK_M + 2 * WARP_SIZE;

    int A_sz = BLOCK_M * BLOCK_K / 2;
    int B_sz = BLOCK_N * BLOCK_K / 2;
    int SF_sz = 128 * BLOCK_K / 16;
    int stage_size = A_sz + B_sz * 2 + SF_sz * 3;
    int smem_size = stage_size * NUM_STAGES;

    auto kernel = dual_gemm_silu_kernel<BLOCK_M, BLOCK_N, BLOCK_K, NUM_STAGES>;
    if (smem_size > 48'000)
        cudaFuncSetAttribute(kernel, cudaFuncAttributeMaxDynamicSharedMemorySize, smem_size);

    kernel<<<grid, tb_size, smem_size>>>(
        A_tmap, B1_tmap, B2_tmap,
        SFA_ptr, SFB1_ptr, SFB2_ptr,
        C_ptr, M, N, K
    );

    return C;
}

TORCH_LIBRARY(dual_gemm_n64, m) {
    m.def("dual_gemm_silu(Tensor A, Tensor B1, Tensor B2, Tensor SFA, Tensor SFB1, Tensor SFB2, Tensor(a!) C) -> Tensor");
    m.impl("dual_gemm_silu", &dual_gemm_silu_n64);
}

// BLOCK_N=128 kernel for m=512 cases (3 stages)
at::Tensor dual_gemm_silu_n128(
    const at::Tensor& A,
    const at::Tensor& B1,
    const at::Tensor& B2,
    const at::Tensor& SFA,
    const at::Tensor& SFB1,
    const at::Tensor& SFB2,
                at::Tensor& C
) {
    const int M = A.size(0);
    const int N = B1.size(0);
    const int K = A.size(1) * 2;

    constexpr int BLOCK_M = 128;
    constexpr int BLOCK_N = 128;
    constexpr int BLOCK_K = 256;
    constexpr int NUM_STAGES = 4;  // 4 stages × 54KB = 216KB (no silu smem needed)

    auto A_ptr   = reinterpret_cast<const char *>(A.data_ptr());
    auto B1_ptr  = reinterpret_cast<const char *>(B1.data_ptr());
    auto B2_ptr  = reinterpret_cast<const char *>(B2.data_ptr());
    auto SFA_ptr = reinterpret_cast<const char *>(SFA.data_ptr());
    auto SFB1_ptr = reinterpret_cast<const char *>(SFB1.data_ptr());
    auto SFB2_ptr = reinterpret_cast<const char *>(SFB2.data_ptr());
    auto C_ptr   = reinterpret_cast<half *>(C.data_ptr());

    CUtensorMap A_tmap, B1_tmap, B2_tmap;
    init_AB_tmap(&A_tmap, A_ptr, M, K, BLOCK_M, BLOCK_K);
    init_AB_tmap(&B1_tmap, B1_ptr, N, K, BLOCK_N, BLOCK_K);
    init_AB_tmap(&B2_tmap, B2_ptr, N, K, BLOCK_N, BLOCK_K);

    int grid = (M / BLOCK_M) * (N / BLOCK_N);
    int tb_size = BLOCK_M + 2 * WARP_SIZE;

    int A_sz = BLOCK_M * BLOCK_K / 2;
    int B_sz = BLOCK_N * BLOCK_K / 2;
    int SF_sz = 128 * BLOCK_K / 16;
    int stage_size = A_sz + B_sz * 2 + SF_sz * 3;
    int smem_size = stage_size * NUM_STAGES;

    auto kernel = dual_gemm_silu_kernel<BLOCK_M, BLOCK_N, BLOCK_K, NUM_STAGES>;
    if (smem_size > 48'000)
        cudaFuncSetAttribute(kernel, cudaFuncAttributeMaxDynamicSharedMemorySize, smem_size);

    kernel<<<grid, tb_size, smem_size>>>(
        A_tmap, B1_tmap, B2_tmap,
        SFA_ptr, SFB1_ptr, SFB2_ptr,
        C_ptr, M, N, K
    );

    return C;
}

TORCH_LIBRARY(dual_gemm_n128, m) {
    m.def("dual_gemm_silu(Tensor A, Tensor B1, Tensor B2, Tensor SFA, Tensor SFB1, Tensor SFB2, Tensor(a!) C) -> Tensor");
    m.impl("dual_gemm_silu", &dual_gemm_silu_n128);
}
"""

# Compile both kernels together so they share common code.
load_inline(
    "dual_gemm_kernels",
    cpp_sources="",
    cuda_sources=cuda_src,
  #  verbose=True,
    is_python_module=False,
    no_implicit_headers=True,
    extra_cuda_cflags=[
        "-O3",
        "-gencode=arch=compute_100a,code=sm_100a",
        "--use_fast_math",
        "--expt-relaxed-constexpr",
        "--relocatable-device-code=false",
        "-lineinfo",
     #   "-Xptxas=-v",
    ],
    extra_ldflags=["-lcuda"],
)

dual_gemm_silu_n64 = torch.ops.dual_gemm_n64.dual_gemm_silu
dual_gemm_silu_n128 = torch.ops.dual_gemm_n128.dual_gemm_silu


def custom_kernel(data: input_t) -> output_t:
    a, b1, b2 = data[0], data[1], data[2]
    sfa_perm, sfb1_perm, sfb2_perm = data[6], data[7], data[8]
    c = data[9]

    M = a.shape[0]
    if M == 256:
        # Use BLOCK_N=64 kernel (5 stages)
        return dual_gemm_silu_n64(a, b1, b2, sfa_perm, sfb1_perm, sfb2_perm, c)
    else:
        # Use BLOCK_N=128 kernel (3 stages) for m=512 cases
        return dual_gemm_silu_n128(a, b1, b2, sfa_perm, sfb1_perm, sfb2_perm, c)