whisper.cpp/ggml/src/ggml-cuda/norm.cu

#include "norm.cuh"
#include <cstdint>

template <int block_size>
static __global__ void norm_f32(
        const float * x, float * dst, const int ncols, const int64_t stride_row, const int64_t stride_channel,
        const int64_t stride_sample, const float eps) {
    const int nrows     = gridDim.x;
    const int nchannels = gridDim.y;

    const int row       = blockIdx.x;
    const int channel   = blockIdx.y;
    const int sample    = blockIdx.z;
    const int tid       = threadIdx.x;

    x   += sample*stride_sample + channel*stride_channel + row*stride_row;
    dst += ((sample*nchannels + channel)*nrows + row)*ncols;

    float2 mean_var = make_float2(0.0f, 0.0f);

    for (int col = tid; col < ncols; col += block_size) {
        const float xi = x[col];
        mean_var.x += xi;
        mean_var.y += xi * xi;
    }

    // sum up partial sums
    mean_var = warp_reduce_sum(mean_var);
    if constexpr (block_size > WARP_SIZE) {
        static_assert(block_size == 1024, "unexpected block_size");
        __shared__ float2 s_sum[32];
        const int warp_id = threadIdx.x / WARP_SIZE;
        const int lane_id = threadIdx.x % WARP_SIZE;
        if (lane_id == 0) {
            s_sum[warp_id] = mean_var;
        }
        __syncthreads();
        mean_var = s_sum[lane_id];
        mean_var = warp_reduce_sum(mean_var);
    }

    const float mean = mean_var.x / ncols;
    const float var = mean_var.y / ncols - mean * mean;
    const float inv_std = rsqrtf(var + eps);

    for (int col = tid; col < ncols; col += block_size) {
        dst[col] = (x[col] - mean) * inv_std;
    }
}

template <int block_size>
static __global__ void group_norm_f32(const float * x, float * dst, const int group_size, const int ne_elements, const float eps) {
    // blockIdx.x: num_groups idx
    // threadIdx.x: block_size idx
    const int start =     blockIdx.x*group_size + threadIdx.x;
    const int end   = min(blockIdx.x*group_size + group_size,  ne_elements);

    float tmp = 0.0f; // partial sum for thread in warp

    for (int j = start; j < end; j += block_size) {
        tmp += x[j];
    }

    tmp = warp_reduce_sum(tmp);
    if constexpr (block_size > WARP_SIZE) {
        static_assert(block_size == 1024, "unexpected block_size");
        __shared__ float s_sum[32];
        const int warp_id = threadIdx.x / WARP_SIZE;
        const int lane_id = threadIdx.x % WARP_SIZE;
        if (lane_id == 0) {
            s_sum[warp_id] = tmp;
        }
        __syncthreads();
        tmp = s_sum[lane_id];
        tmp = warp_reduce_sum(tmp);
    }

    const float mean = tmp / group_size;
    tmp = 0.0f;

    for (int j = start; j < end; j += block_size) {
        const float xi = x[j] - mean;
        dst[j] = xi;
        tmp += xi * xi;
    }

    tmp = warp_reduce_sum(tmp);
    if (block_size > WARP_SIZE) {
        __shared__ float s_sum[32];
        const int warp_id = threadIdx.x / WARP_SIZE;
        const int lane_id = threadIdx.x % WARP_SIZE;
        if (lane_id == 0) {
            s_sum[warp_id] = tmp;
        }
        __syncthreads();
        tmp = s_sum[lane_id];
        tmp = warp_reduce_sum(tmp);
    }

    const float variance = tmp / group_size;
    const float scale = rsqrtf(variance + eps);
    for (int j = start; j < end; j += block_size) {
        dst[j] *= scale;
    }
}

template <int block_size>
static __global__ void rms_norm_f32(
        const float * x, float * dst, const int ncols, const int64_t stride_row, const int64_t stride_channel,
        const int64_t stride_sample, const float eps) {
    const int nrows     = gridDim.x;
    const int nchannels = gridDim.y;

    const int row       = blockIdx.x;
    const int channel   = blockIdx.y;
    const int sample    = blockIdx.z;
    const int tid       = threadIdx.x;

    x   += sample*stride_sample + channel*stride_channel + row*stride_row;
    dst += ((sample*nchannels + channel)*nrows + row)*ncols;

    float tmp = 0.0f; // partial sum for thread in warp

    for (int col = tid; col < ncols; col += block_size) {
        const float xi = x[col];
        tmp += xi * xi;
    }

    // sum up partial sums
    tmp = warp_reduce_sum(tmp);
    if constexpr (block_size > WARP_SIZE) {
        static_assert(block_size == 1024, "unexpected block_size");
        __shared__ float s_sum[32];
        const int warp_id = threadIdx.x / WARP_SIZE;
        const int lane_id = threadIdx.x % WARP_SIZE;
        if (lane_id == 0) {
            s_sum[warp_id] = tmp;
        }
        __syncthreads();
        tmp = s_sum[lane_id];
        tmp = warp_reduce_sum(tmp);
    }

    const float mean = tmp / ncols;
    const float scale = rsqrtf(mean + eps);

    for (int col = tid; col < ncols; col += block_size) {
        dst[col] = scale * x[col];
    }
}

template <int block_size>
static __global__ void rms_norm_back_f32(
        const float * grad, const float * xf, float * dst, const int ncols, const float eps) {
    const int row = blockIdx.x*blockDim.y + threadIdx.y;
    const int tid = threadIdx.x;

    grad += int64_t(row)*ncols;
    xf   += int64_t(row)*ncols;
    dst  += int64_t(row)*ncols;

    float sum_xx = 0.0f; // sum for squares of x, equivalent to forward pass
    float sum_xg = 0.0f; // sum for x * gradient, needed because RMS norm mixes inputs

    for (int col = tid; col < ncols; col += block_size) {
        const float xfi = xf[col];
        sum_xx += xfi * xfi;
        sum_xg += xfi * grad[col];
    }

    // sum up partial sums
    sum_xx = warp_reduce_sum(sum_xx);
    sum_xg = warp_reduce_sum(sum_xg);
    if constexpr (block_size > WARP_SIZE) {
        static_assert(block_size == 1024, "unexpected block_size");
        __shared__ float s_sum_xx[32];
        __shared__ float s_sum_xg[32];
        const int warp_id = threadIdx.x / WARP_SIZE;
        const int lane_id = threadIdx.x % WARP_SIZE;
        if (lane_id == 0) {
            s_sum_xx[warp_id] = sum_xx;
            s_sum_xg[warp_id] = sum_xg;
        }
        __syncthreads();

        sum_xx = s_sum_xx[lane_id];
        sum_xx = warp_reduce_sum(sum_xx);

        sum_xg = s_sum_xg[lane_id];
        sum_xg = warp_reduce_sum(sum_xg);
    }

    const float mean_eps = sum_xx / ncols + eps;
    const float sum_eps  = sum_xx + ncols*eps;

    const float scale_grad = rsqrtf(mean_eps);
    const float scale_x    = -scale_grad * sum_xg/sum_eps;

    for (int col = tid; col < ncols; col += block_size) {
        dst[col] = scale_grad*grad[col] + scale_x*xf[col];
    }
}

// template <int block_size>
// static __global__ void l2_norm_f32(const float * x, float * dst, const int ncols, const float eps) {
//     const int row = blockIdx.x*blockDim.y + threadIdx.y;
//     const int tid = threadIdx.x;

//     float tmp = 0.0f; // partial sum for thread in warp

//     for (int col = tid; col < ncols; col += block_size) {
//         const float xi = x[row*ncols + col];
//         tmp += xi * xi;
//     }

//     // sum up partial sums
//     tmp = warp_reduce_sum(tmp);
//     if (block_size > WARP_SIZE) {
//         __shared__ float s_sum[32];
//         int warp_id = threadIdx.x / WARP_SIZE;
//         int lane_id = threadIdx.x % WARP_SIZE;
//         if (lane_id == 0) {
//             s_sum[warp_id] = tmp;
//         }
//         __syncthreads();
//         tmp = s_sum[lane_id];
//         tmp = warp_reduce_sum(tmp);
//     }

//     // from https://pytorch.org/docs/stable/generated/torch.nn.functional.normalize.html
//     const float scale = rsqrtf(fmaxf(tmp, eps * eps));

//     for (int col = tid; col < ncols; col += block_size) {
//         dst[row*ncols + col] = scale * x[row*ncols + col];
//     }
// }

template <int block_size>
static __global__ void l2_norm_f32(
        const float * x, float * dst, const int ncols, const int64_t stride_row, const int64_t stride_channel,
        const int64_t stride_sample, const float eps) {
    const int nrows     = gridDim.x;
    const int nchannels = gridDim.y;

    const int row       = blockIdx.x;
    const int channel   = blockIdx.y;
    const int sample    = blockIdx.z;
    const int tid       = threadIdx.x;

    x   += sample*stride_sample + channel*stride_channel + row*stride_row;
    dst += ((sample*nchannels + channel)*nrows + row)*ncols;

    float tmp = 0.0f; // partial sum for thread in warp

    for (int col = tid; col < ncols; col += block_size) {
        const float xi = x[col];
        tmp += xi * xi;
    }

    // sum up partial sums
    tmp = warp_reduce_sum(tmp);
    if constexpr (block_size > WARP_SIZE) {
        static_assert(block_size == 1024, "unexpected block_size");
        __shared__ float s_sum[32];
        const int warp_id = threadIdx.x / WARP_SIZE;
        const int lane_id = threadIdx.x % WARP_SIZE;
        if (lane_id == 0) {
            s_sum[warp_id] = tmp;
        }
        __syncthreads();
        tmp = s_sum[lane_id];
        tmp = warp_reduce_sum(tmp);
    }

    // from https://pytorch.org/docs/stable/generated/torch.nn.functional.normalize.html
    const float scale = rsqrtf(fmaxf(tmp, eps * eps));

    for (int col = tid; col < ncols; col += block_size) {
        dst[col] = scale * x[col];
    }
}

static void norm_f32_cuda(
        const float * x, float * dst, const int ncols, const int nrows, const int nchannels, const int nsamples,
        const int64_t stride_row, const int64_t stride_channel, const int64_t stride_sample, const float eps, cudaStream_t stream) {
    const dim3 blocks_num(nrows, nchannels, nsamples);
    if (ncols < 1024) {
        const dim3 block_dims(WARP_SIZE, 1, 1);
        norm_f32<WARP_SIZE><<<blocks_num, block_dims, 0, stream>>>(x, dst, ncols, stride_row, stride_channel, stride_sample, eps);
    } else {
        const dim3 block_dims(1024, 1, 1);
        norm_f32<1024><<<blocks_num, block_dims, 0, stream>>>(x, dst, ncols, stride_row, stride_channel, stride_sample, eps);
    }
}

static void group_norm_f32_cuda(
        const float * x, float * dst, const int num_groups, const float eps, const int group_size, const int ne_elements, cudaStream_t stream) {
    if (group_size < 1024) {
        const dim3 block_dims(WARP_SIZE, 1, 1);
        group_norm_f32<WARP_SIZE><<<num_groups, block_dims, 0, stream>>>(x, dst, group_size, ne_elements, eps);
    } else {
        const dim3 block_dims(1024, 1, 1);
        group_norm_f32<1024><<<num_groups, block_dims, 0, stream>>>(x, dst, group_size, ne_elements, eps);
    }
}

static void rms_norm_f32_cuda(
        const float * x, float * dst, const int ncols, const int nrows, const int nchannels, const int nsamples,
        const int64_t stride_row, const int64_t stride_channel, const int64_t stride_sample, const float eps, cudaStream_t stream) {
    const dim3 blocks_num(nrows, nchannels, nsamples);
    if (ncols < 1024) {
        const dim3 block_dims(WARP_SIZE, 1, 1);
        rms_norm_f32<WARP_SIZE><<<blocks_num, block_dims, 0, stream>>>(x, dst, ncols, stride_row, stride_channel, stride_sample, eps);
    } else {
        const dim3 block_dims(1024, 1, 1);
        rms_norm_f32<1024><<<blocks_num, block_dims, 0, stream>>>(x, dst, ncols, stride_row, stride_channel, stride_sample, eps);
    }
}

static void rms_norm_back_f32_cuda(const float * grad, const float * xf, float * dst, const int ncols, const int nrows, const float eps, cudaStream_t stream) {
    if (ncols < 1024) {
        const dim3 block_dims(WARP_SIZE, 1, 1);
        rms_norm_back_f32<WARP_SIZE><<<nrows, block_dims, 0, stream>>>(grad, xf, dst, ncols, eps);
    } else {
        const dim3 block_dims(1024, 1, 1);
        rms_norm_back_f32<1024><<<nrows, block_dims, 0, stream>>>(grad, xf, dst, ncols, eps);
    }
}

static void l2_norm_f32_cuda(
        const float * x, float * dst, const int ncols, const int nrows, const int nchannels, const int nsamples,
        const int64_t stride_row, const int64_t stride_channel, const int64_t stride_sample, const float eps, cudaStream_t stream) {
    const dim3 blocks_num(nrows, nchannels, nsamples);
    if (ncols < 1024) {
        const dim3 block_dims(WARP_SIZE, 1, 1);
        l2_norm_f32<WARP_SIZE><<<blocks_num, block_dims, 0, stream>>>(x, dst, ncols, stride_row, stride_channel, stride_sample, eps);
    } else {
        const dim3 block_dims(1024, 1, 1);
        l2_norm_f32<1024><<<blocks_num, block_dims, 0, stream>>>(x, dst, ncols, stride_row, stride_channel, stride_sample, eps);
    }
}

void ggml_cuda_op_norm(ggml_backend_cuda_context & ctx, ggml_tensor * dst) {
    const ggml_tensor * src0 = dst->src[0];
    const float * src0_d = (const float *) src0->data;
    float * dst_d = (float *) dst->data;
    cudaStream_t stream = ctx.stream();

    GGML_ASSERT(src0->type == GGML_TYPE_F32);
    GGML_ASSERT( dst->type == GGML_TYPE_F32);

    GGML_TENSOR_UNARY_OP_LOCALS;

    float eps;
    memcpy(&eps, dst->op_params, sizeof(float));
    GGML_ASSERT(eps >= 0.0f);

    const size_t ts0 = ggml_type_size(src0->type);
    GGML_ASSERT(nb00 == ts0);
    const int64_t s01 = nb01 / ts0;
    const int64_t s02 = nb02 / ts0;
    const int64_t s03 = nb03 / ts0;

    norm_f32_cuda(src0_d, dst_d, ne00, ne01, ne02, ne03, s01, s02, s03, eps, stream);
}

void ggml_cuda_op_group_norm(ggml_backend_cuda_context & ctx, ggml_tensor * dst) {
    const ggml_tensor * src0 = dst->src[0];
    const float * src0_d = (const float *)src0->data;
    float * dst_d = (float *)dst->data;
    cudaStream_t stream = ctx.stream();

    GGML_ASSERT(src0->type == GGML_TYPE_F32);
    GGML_ASSERT( dst->type == GGML_TYPE_F32);

    int num_groups = dst->op_params[0];

    float eps;
    memcpy(&eps, dst->op_params + 1, sizeof(float));
    GGML_ASSERT(eps >= 0.0f);

    int group_size = src0->ne[0] * src0->ne[1] * ((src0->ne[2] + num_groups - 1) / num_groups);
    group_norm_f32_cuda(src0_d, dst_d, num_groups * src0->ne[3], eps, group_size, ggml_nelements(src0), stream);
}

void ggml_cuda_op_rms_norm(ggml_backend_cuda_context & ctx, ggml_tensor * dst) {
    const ggml_tensor * src0 = dst->src[0];
    const float * src0_d = (const float *) src0->data;
    float * dst_d = (float *) dst->data;
    cudaStream_t stream = ctx.stream();

    GGML_ASSERT(src0->type == GGML_TYPE_F32);
    GGML_ASSERT( dst->type == GGML_TYPE_F32);

    GGML_TENSOR_UNARY_OP_LOCALS;

    float eps;
    memcpy(&eps, dst->op_params, sizeof(float));
    GGML_ASSERT(eps >= 0.0f);

    const size_t ts0 = ggml_type_size(src0->type);
    GGML_ASSERT(nb00 == ts0);
    const int64_t s01 = nb01 / ts0;
    const int64_t s02 = nb02 / ts0;
    const int64_t s03 = nb03 / ts0;

    rms_norm_f32_cuda(src0_d, dst_d, ne00, ne01, ne02, ne03, s01, s02, s03, eps, stream);
}

void ggml_cuda_op_rms_norm_back(ggml_backend_cuda_context & ctx, ggml_tensor * dst) {
    const ggml_tensor * grad  = dst->src[0]; // gradients
    const ggml_tensor * src0f = dst->src[1]; // src0 from forward pass

    const float * grad_d  = (const float *) grad->data;
    const float * src0f_d = (const float *) src0f->data;
    float       * dst_d   = (float       *) dst->data;

    cudaStream_t stream = ctx.stream();

    GGML_ASSERT(ggml_is_contiguous(grad));

    GGML_ASSERT( grad->type == GGML_TYPE_F32);
    GGML_ASSERT(src0f->type == GGML_TYPE_F32);
    GGML_ASSERT(  dst->type == GGML_TYPE_F32);

    const int64_t ne00 = src0f->ne[0];
    const int64_t nrows = ggml_nrows(src0f);

    float eps;
    memcpy(&eps, dst->op_params, sizeof(float));
    GGML_ASSERT(eps >= 0.0f);

    rms_norm_back_f32_cuda(grad_d, src0f_d, dst_d, ne00, nrows, eps, stream);
}

void ggml_cuda_op_l2_norm(ggml_backend_cuda_context & ctx, ggml_tensor * dst) {
    const ggml_tensor * src0 = dst->src[0];
    const float * src0_d = (const float *) src0->data;
    float * dst_d = (float *) dst->data;
    cudaStream_t stream = ctx.stream();

    GGML_ASSERT(src0->type == GGML_TYPE_F32);
    GGML_ASSERT( dst->type == GGML_TYPE_F32);

    GGML_TENSOR_UNARY_OP_LOCALS;

    float eps;
    memcpy(&eps, dst->op_params, sizeof(float));
    GGML_ASSERT(eps >= 0.0f);

    const size_t ts0 = ggml_type_size(src0->type);
    GGML_ASSERT(nb00 == ts0);
    const int64_t s01 = nb01 / ts0;
    const int64_t s02 = nb02 / ts0;
    const int64_t s03 = nb03 / ts0;

    l2_norm_f32_cuda(src0_d, dst_d, ne00, ne01, ne02, ne03, s01, s02, s03, eps, stream);
}