whispercpp 1.3.2 → 1.3.4

data/ext/sources/ggml/src/ggml-cuda/fattn-tile.cuh

@@ -0,0 +1,3 @@
+#include "common.cuh"
+
+void ggml_cuda_flash_attn_ext_tile(ggml_backend_cuda_context & ctx, ggml_tensor * dst);

data/ext/sources/ggml/src/ggml-cuda/fattn-vec.cuh

@@ -0,0 +1,593 @@
+#include "common.cuh"
+#include "fattn-common.cuh"
+
+static int ggml_cuda_fattn_vec_get_nthreads_host(const int cc) {
+    return 128;
+    GGML_UNUSED(cc);
+}
+
+static constexpr __device__ int ggml_cuda_fattn_vec_get_nthreads_device() {
+    return 128;
+}
+
+// Currenlty llvm with the amdgcn target dose not support unrolling loops
+// that contain a break that can not be resolved at compile time.
+#ifdef __clang__
+#pragma clang diagnostic push
+#pragma clang diagnostic ignored "-Wpass-failed"
+#endif // __clang__
+template<int D, int ncols, ggml_type type_K, ggml_type type_V, bool use_logit_softcap> // D == head size
+__launch_bounds__(ggml_cuda_fattn_vec_get_nthreads_device(), 1)
+static __global__ void flash_attn_ext_vec(
+        const char * __restrict__ Q,
+        const char * __restrict__ K,
+        const char * __restrict__ V,
+        const char * __restrict__ mask,
+        const char * __restrict__ sinks,
+        const int  * __restrict__ KV_max,
+        float      * __restrict__ dst,
+        float2     * __restrict__ dst_meta,
+        const float scale,
+        const float max_bias,
+        const float m0,
+        const float m1,
+        const uint32_t n_head_log2,
+        const float logit_softcap,
+        const int32_t ne00, const int32_t ne01, const int32_t ne02, const int32_t ne03,
+                            const int32_t nb01, const int32_t nb02, const int32_t nb03,
+        const int32_t ne10, const int32_t ne11, const int32_t ne12, const int32_t ne13,
+                            const int32_t nb11, const int32_t nb12, const int64_t nb13,
+                            const int32_t nb21, const int32_t nb22, const int64_t nb23,
+                            const int32_t ne31, const int32_t ne32, const int32_t ne33,
+                            const int32_t nb31, const int32_t nb32, const int64_t nb33) {
+#ifdef FLASH_ATTN_AVAILABLE
+
+    // Skip unused kernel variants for faster compilation:
+    if (use_logit_softcap && !(D == 128 || D == 256)) {
+        GGML_UNUSED_VARS(Q, K, V, mask, sinks, KV_max, dst, dst_meta, scale,
+            max_bias, m0, m1, n_head_log2, logit_softcap,
+            ne00, ne01, ne02, ne03,
+                  nb01, nb02, nb03,
+            ne10, ne11, ne12, ne13,
+                  nb11, nb12, nb13,
+                  nb21, nb22, nb23,
+                  ne31, ne32, ne33,
+                  nb31, nb32, nb33);
+        NO_DEVICE_CODE;
+        return;
+    }
+
+    //In this kernel Q, K, V are matrices while i, j, k are matrix indices.
+
+    constexpr int cpy_nb = ggml_cuda_get_max_cpy_bytes();
+    constexpr int cpy_ne = cpy_nb / 4;
+
+#ifdef GGML_USE_HIP
+#ifdef RDNA
+    constexpr int nthreads_KQ_q = 2;
+#else
+    constexpr int nthreads_KQ_q = 4;
+#endif // RDNA
+    constexpr int nthreads_V_q  = (D/4 < 32 ? D/4 : 32);
+#else
+    constexpr int nthreads_KQ_q = (D/4 < 32 ? D/4 : 32);
+    constexpr int nthreads_V_q  = (D/4 < 32 ? D/4 : 32);
+#endif // GGML_USE_HIP
+
+    constexpr int nthreads    = ggml_cuda_fattn_vec_get_nthreads_device();
+    constexpr int nthreads_KQ = type_K == GGML_TYPE_F16 ? 128 / cpy_nb : nthreads_KQ_q;
+    constexpr int nthreads_V  = type_V == GGML_TYPE_F16 ? 128 / cpy_nb : nthreads_V_q;
+
+    static_assert(WARP_SIZE % nthreads_KQ == 0, "bad nthreads_K");
+    static_assert(WARP_SIZE % nthreads_V  == 0, "bad nthreads_V");
+
+    constexpr int V_rows_per_thread = type_V == GGML_TYPE_F16 ? 2*cpy_ne : 4;
+    constexpr int V_cols_per_iter   = WARP_SIZE / nthreads_V;
+
+    constexpr vec_dot_KQ_t vec_dot_KQ = get_vec_dot_KQ<type_K, D, nthreads_KQ>();
+    constexpr bool Q_q8_1 = type_K != GGML_TYPE_F16;
+#ifdef FAST_FP16_AVAILABLE
+    constexpr dequantize_V_t dequantize_V = get_dequantize_V<type_V, half,  V_rows_per_thread>();
+#else
+    constexpr dequantize_V_t dequantize_V = get_dequantize_V<type_V, float, V_rows_per_thread>();
+#endif // FAST_FP16_AVAILABLE
+
+    const int ic0 = blockIdx.x * ncols; // Index of the Q/QKV column to work on.
+
+    const int sequence = blockIdx.z / ne02;
+    const int head = blockIdx.z - sequence*ne02;
+    const int gqa_ratio = ne02 / ne12; // With grouped query attention there are > 1 Q matrices per K, V matrix.
+    Q += nb03*sequence + nb02* head              + nb01*ic0;
+    K += nb13*sequence + nb12*(head / gqa_ratio);
+    V += nb23*sequence + nb22*(head / gqa_ratio);
+
+    const half * maskh  = (const half  *) (mask + nb33*(sequence % ne33) + nb31*ic0);
+
+    const float slope = get_alibi_slope(max_bias, head, n_head_log2, m0, m1);
+
+    static_assert(D % (2*WARP_SIZE) == 0, "D not divisible by 2*WARP_SIZE == 64.");
+    constexpr int nwarps = nthreads / WARP_SIZE;
+    const int tid = WARP_SIZE*threadIdx.y + threadIdx.x;
+    __builtin_assume(tid < nthreads);
+
+    constexpr int ne_KQ      = ncols*D;
+    constexpr int ne_combine = nwarps*V_cols_per_iter*D;
+#ifdef FAST_FP16_AVAILABLE
+    half2            VKQ[ncols][(D/2)/nthreads_V] = {{{0.0f, 0.0f}}};
+    __shared__ half   KQ[ne_KQ > ne_combine ? ne_KQ : ne_combine];
+#else
+    float2           VKQ[ncols][(D/2)/nthreads_V] = {{{0.0f, 0.0f}}};
+    __shared__ float  KQ[ne_KQ > ne_combine ? ne_KQ : ne_combine];
+#endif // FAST_FP16_AVAILABLE
+
+    float KQ_max[ncols];
+    float KQ_sum[ncols];
+#pragma unroll
+    for (int j = 0; j < ncols; ++j) {
+        KQ_max[j] = -FLT_MAX/2.0f;
+        KQ_sum[j] = 0.0f;
+    }
+
+    // Convert Q to float2 (f16 K) or q8_1 (quantized K) and store in registers:
+#ifdef FAST_FP16_AVAILABLE
+    half2  Q_reg[ncols][(D/2)/nthreads_KQ]; // Will be initialized completely.
+#else
+    float2 Q_reg[ncols][(D/2)/nthreads_KQ] = {{{0.0f, 0.0f}}}; // May be only partially initialized.
+#endif // FAST_FP16_AVAILABLE
+    int    Q_i32[ncols][1 > D/(sizeof(int)*nthreads_KQ) ? 1 : D/(sizeof(int)*nthreads_KQ)];
+    float2  Q_ds[ncols][1 > D/(sizeof(int)*nthreads_KQ) ? 1 : D/(sizeof(int)*nthreads_KQ)];
+    if constexpr (Q_q8_1) {
+#pragma unroll
+        for (int j0 = 0; j0 < ncols; j0 += nwarps) {
+            const int j = j0 + threadIdx.y;
+
+            if (j0 + nwarps > ncols && j >= ncols) {
+                break;
+            }
+
+            // Reuse KQ as temporary storage for converting Q to q8_1:
+            int    * tmp_q_i32 = (int    *) &KQ[j*D];
+            float2 * tmp_q_ds  = (float2 *) (tmp_q_i32 + D/sizeof(int));
+
+            // Set memory to zero if out of bounds:
+            if (ncols > 1 && ic0 + j >= ne01) {
+#pragma unroll
+                for (int i0 = 0; i0 < int(D/sizeof(int)); i0 += WARP_SIZE) {
+                    const int i = i0 + threadIdx.x;
+
+                    if (i0 + WARP_SIZE <= D/sizeof(int) || i < D/sizeof(int)) {
+                        tmp_q_i32[i] = 0;
+                    }
+                }
+                if (threadIdx.x < D/QK8_1) {
+                    tmp_q_ds[threadIdx.x] = make_float2(0.0f, 0.0f);
+                }
+            } else {
+                const float * Q_f = (const float *) (Q + j*nb01);
+                constexpr int nthreads_quantize = D/sizeof(int) < WARP_SIZE ? D/sizeof(int) : WARP_SIZE;
+#pragma unroll
+                for (int i0 = 0; i0 < int(D/sizeof(int)); i0 += nthreads_quantize) {
+                    quantize_q8_1_to_shared<float2, nthreads_quantize>
+                        (Q_f + i0*sizeof(int), scale, tmp_q_i32 + i0, tmp_q_ds + i0/QI8_1);
+                }
+            }
+        }
+
+        __syncthreads();
+
+#pragma unroll
+        for (int j = 0; j < ncols; ++j) {
+            int    * tmp_q_i32 = (int    *) &KQ[j*D];
+            float2 * tmp_q_ds  = (float2 *) (tmp_q_i32 + D/sizeof(int));
+
+#pragma unroll
+            for (int i0 = 0; i0 < int(D/sizeof(int)); i0 += nthreads_KQ) {
+                const int i = i0 + (nthreads_KQ == WARP_SIZE ? threadIdx.x : threadIdx.x % nthreads_KQ);
+
+                Q_i32[j][i0/nthreads_KQ] = tmp_q_i32[i];
+                Q_ds[j][i0/nthreads_KQ]  = tmp_q_ds[i/QI8_1];
+            }
+        }
+
+        __syncthreads();
+    } else {
+#ifdef FAST_FP16_AVAILABLE
+        const half2 scale_h2 = make_half2(scale, scale);
+#pragma unroll
+        for (int j = 0; j < ncols; ++j) {
+            const float2 * Q_j = (const float2 *) (Q + j*nb01);
+#pragma unroll
+            for (int i0 = 0; i0 < D/2; i0 += nthreads_KQ*cpy_ne) {
+                const int i = i0 + (nthreads_KQ == WARP_SIZE ? threadIdx.x : threadIdx.x % nthreads_KQ)*cpy_ne;
+
+                float2 tmp[cpy_ne] = {{0.0f, 0.0f}};
+                if (ncols == 1 || ic0 + j < ne01) {
+                    ggml_cuda_memcpy_1<cpy_nb>(tmp,            &Q_j[i]);
+                    ggml_cuda_memcpy_1<cpy_nb>(tmp + cpy_ne/2, &Q_j[i + cpy_ne/2]);
+                }
+#pragma unroll
+                for (int i1 = 0; i1 < cpy_ne; ++i1) {
+                    Q_reg[j][i0/nthreads_KQ + i1] = make_half2(tmp[i1].x, tmp[i1].y);
+                }
+            }
+#pragma unroll
+            for (int k = 0; k < (D/2)/nthreads_KQ; ++k) {
+                Q_reg[j][k] *= scale_h2;
+            }
+        }
+#else
+#pragma unroll
+        for (int j = 0; j < ncols; ++j) {
+            const float2 * Q_j = (const float2 *) (Q + j*nb01);
+#pragma unroll
+            for (int i0 = 0; i0 < D/2; i0 += nthreads_KQ*cpy_ne) {
+                const int i = i0 + (nthreads_KQ == WARP_SIZE ? threadIdx.x : threadIdx.x % nthreads_KQ)*cpy_ne;
+                if (ncols == 1 || ic0 + j < ne01) {
+                    ggml_cuda_memcpy_1<cpy_nb>(&Q_reg[j][i0/nthreads_KQ],            &Q_j[i]);
+                    ggml_cuda_memcpy_1<cpy_nb>(&Q_reg[j][i0/nthreads_KQ + cpy_ne/2], &Q_j[i + cpy_ne/2]);
+                }
+            }
+#pragma unroll
+            for (int k = 0; k < (D/2)/nthreads_KQ; ++k) {
+                Q_reg[j][k].x *= scale;
+                Q_reg[j][k].y *= scale;
+            }
+        }
+#endif // FAST_FP16_AVAILABLE
+    }
+
+    const int k_VKQ_max = KV_max ? KV_max[sequence*gridDim.x + blockIdx.x] : ne11;
+    K     += blockIdx.y*nthreads * nb11;
+    V     += blockIdx.y*nthreads * nb21;
+    maskh += blockIdx.y*nthreads;
+    for (int k_VKQ_0 = blockIdx.y*nthreads; k_VKQ_0 < k_VKQ_max; k_VKQ_0 += gridDim.y*nthreads,
+             // Increment pointers after each loop:
+             K += gridDim.y*nthreads*nb11, V += gridDim.y*nthreads*nb21, maskh += gridDim.y*nthreads) {
+
+        // Calculate KQ tile and keep track of new maximum KQ values:
+        float KQ_reg[ncols]; // KQ in registers.
+
+        float KQ_max_new[ncols];
+#pragma unroll
+        for (int j = 0; j < ncols; ++j) {
+            KQ_max_new[j] = KQ_max[j];
+        }
+
+#pragma unroll
+        for (int i_KQ_0 = 0; i_KQ_0 < nthreads_KQ; ++i_KQ_0) {
+            const int i_KQ = threadIdx.y*WARP_SIZE + (nthreads_KQ == WARP_SIZE ? 0 : (threadIdx.x & ~(nthreads_KQ-1))) + i_KQ_0;
+
+#pragma unroll
+            for (int j = 0; j < ncols; ++j) {
+                float sum = vec_dot_KQ(K + i_KQ*nb11, Q_reg[j], Q_i32[j], Q_ds[j]);
+                sum = warp_reduce_sum<nthreads_KQ>(sum);
+
+                if (use_logit_softcap) {
+                    sum = logit_softcap*tanhf(sum);
+                }
+
+                if (mask) {
+                    sum += slope*__half2float(maskh[j*ne11 + i_KQ]);
+                }
+
+                KQ_max_new[j] = fmaxf(KQ_max_new[j], sum);
+
+                if ((nthreads_KQ == WARP_SIZE ? threadIdx.x : threadIdx.x % nthreads_KQ) == i_KQ_0) {
+                    KQ_reg[j] = sum;
+                }
+            }
+        }
+
+#pragma unroll
+        for (int j = 0; j < ncols; ++j) {
+#pragma unroll
+            for (int offset = nthreads_KQ; offset < WARP_SIZE; offset <<= 1) {
+                KQ_max_new[j] = fmaxf(KQ_max_new[j], __shfl_xor_sync(0xFFFFFFFF, KQ_max_new[j], offset, WARP_SIZE));
+            }
+            const float KQ_max_scale = expf(KQ_max[j] - KQ_max_new[j]);
+            KQ_max[j] = KQ_max_new[j];
+
+            KQ_reg[j] = expf(KQ_reg[j] - KQ_max[j]);
+            KQ_sum[j] = KQ_sum[j]*KQ_max_scale + KQ_reg[j];
+            KQ[j*nthreads + tid] = KQ_reg[j];
+
+#ifdef FAST_FP16_AVAILABLE
+            const half2 KQ_max_scale_h2 = make_half2(KQ_max_scale, KQ_max_scale);
+#pragma unroll
+            for (int i_VKQ_0 = 0; i_VKQ_0 < D/2; i_VKQ_0 += nthreads_V) {
+                VKQ[j][i_VKQ_0/nthreads_V] *= KQ_max_scale_h2;
+            }
+#else
+#pragma unroll
+            for (int i_VKQ_0 = 0; i_VKQ_0 < D/2; i_VKQ_0 += nthreads_V) {
+                VKQ[j][i_VKQ_0/nthreads_V].x *= KQ_max_scale;
+                VKQ[j][i_VKQ_0/nthreads_V].y *= KQ_max_scale;
+            }
+#endif // FAST_FP16_AVAILABLE
+        }
+
+#ifndef GGML_USE_HIP
+        __syncwarp();
+#endif // GGML_USE_HIP
+
+#pragma unroll
+        for (int k0 = 0; k0 < WARP_SIZE; k0 += V_cols_per_iter) {
+            const int k = threadIdx.y*WARP_SIZE + k0 + (nthreads_V == WARP_SIZE ? 0 : threadIdx.x / nthreads_V);
+
+#ifdef FAST_FP16_AVAILABLE
+            half2 KQ_k[ncols];
+#pragma unroll
+            for (int j = 0; j < ncols; ++j) {
+                KQ_k[j] = __half2half2(KQ[j*nthreads + k]);
+            }
+#pragma unroll
+            for (int i_VKQ_0 = 0; i_VKQ_0 < D/2; i_VKQ_0 += nthreads_V*V_rows_per_thread/2) {
+                half2 tmp[V_rows_per_thread/2];
+                dequantize_V(V + k*nb21, tmp,
+                    2*i_VKQ_0 + (nthreads_V == WARP_SIZE ? threadIdx.x : threadIdx.x % nthreads_V)*V_rows_per_thread);
+#pragma unroll
+                for (int i_VKQ_1 = 0; i_VKQ_1 < V_rows_per_thread/2; ++i_VKQ_1) {
+#pragma unroll
+                    for (int j = 0; j < ncols; ++j) {
+                        VKQ[j][i_VKQ_0/nthreads_V + i_VKQ_1] += tmp[i_VKQ_1]*KQ_k[j];
+                    }
+                }
+            }
+#else
+            float KQ_k[ncols];
+#pragma unroll
+            for (int j = 0; j < ncols; ++j) {
+                KQ_k[j] = KQ[j*nthreads + k];
+            }
+#pragma unroll
+            for (int i_VKQ_0 = 0; i_VKQ_0 < D/2; i_VKQ_0 += nthreads_V*V_rows_per_thread/2) {
+                float2 tmp[V_rows_per_thread/2];
+                dequantize_V(V + k*nb21, tmp,
+                    2*i_VKQ_0 + (nthreads_V == WARP_SIZE ? threadIdx.x : threadIdx.x % nthreads_V)*V_rows_per_thread);
+#pragma unroll
+                for (int i_VKQ_1 = 0; i_VKQ_1 < V_rows_per_thread/2; ++i_VKQ_1) {
+#pragma unroll
+                    for (int j = 0; j < ncols; ++j) {
+                        VKQ[j][i_VKQ_0/nthreads_V + i_VKQ_1].x += tmp[i_VKQ_1].x*KQ_k[j];
+                        VKQ[j][i_VKQ_0/nthreads_V + i_VKQ_1].y += tmp[i_VKQ_1].y*KQ_k[j];
+                    }
+                }
+            }
+#endif // FAST_FP16_AVAILABLE
+        }
+    }
+
+    if (sinks && blockIdx.y == 0) {
+        const float sink = ((const float *) sinks)[head];
+
+#pragma unroll
+        for (int j0 = 0; j0 < ncols; j0 += nwarps) {
+            const int j = j0 + threadIdx.y;
+
+            if (j0 + nwarps > ncols && j >= ncols) {
+                break;
+            }
+
+            const float kqmax_new_j = fmaxf(sink, KQ_max[j]);
+            const float KQ_max_scale = expf(KQ_max[j] - kqmax_new_j);
+            KQ_max[j] = kqmax_new_j;
+
+            KQ_sum[j] = KQ_sum[j]*KQ_max_scale + (threadIdx.x == 0 ? expf(sink - KQ_max[j]) : 0.0f);
+
+#ifdef FAST_FP16_AVAILABLE
+            const half2 KQ_max_scale_h2 = make_half2(KQ_max_scale, KQ_max_scale);
+#pragma unroll
+            for (int i_VKQ_0 = 0; i_VKQ_0 < D/2; i_VKQ_0 += nthreads_V) {
+                VKQ[j][i_VKQ_0/nthreads_V] *= KQ_max_scale_h2;
+            }
+#else
+#pragma unroll
+            for (int i_VKQ_0 = 0; i_VKQ_0 < D/2; i_VKQ_0 += nthreads_V) {
+                VKQ[j][i_VKQ_0/nthreads_V].x *= KQ_max_scale;
+                VKQ[j][i_VKQ_0/nthreads_V].y *= KQ_max_scale;
+            }
+#endif // FAST_FP16_AVAILABLE
+        }
+    }
+
+    __shared__ float KQ_max_shared[ncols][WARP_SIZE];
+    __shared__ float KQ_sum_shared[ncols][WARP_SIZE];
+#pragma unroll
+    for (int j = 0; j < ncols; ++j) {
+        if (threadIdx.y == 0) {
+            KQ_max_shared[j][threadIdx.x] = -FLT_MAX/2.0f;
+            KQ_sum_shared[j][threadIdx.x] = 0.0f;
+        }
+    }
+
+    __syncthreads();
+
+#pragma unroll
+    for (int j = 0; j < ncols; ++j) {
+        if (threadIdx.x == 0) {
+            KQ_max_shared[j][threadIdx.y] = KQ_max[j];
+        }
+    }
+    __syncthreads();
+
+#pragma unroll
+    for (int j_VKQ = 0; j_VKQ < ncols; ++j_VKQ) {
+        if (ncols > 1 && ic0 + j_VKQ >= ne01) {
+            break;
+        }
+
+        float kqmax_new = KQ_max_shared[j_VKQ][threadIdx.x];
+        kqmax_new = warp_reduce_max(kqmax_new);
+        const float kqmax_scale = expf(KQ_max[j_VKQ] - kqmax_new);
+        KQ_max[j_VKQ] = kqmax_new;
+
+#ifdef FAST_FP16_AVAILABLE
+        half2 * VKQ_tmp = (half2 *) KQ + threadIdx.y*(V_cols_per_iter*D/2)
+            + (nthreads_V == WARP_SIZE ? 0 : threadIdx.x / nthreads_V)*(D/2);
+
+        const half2 kqmax_scale_h2 = make_half2(kqmax_scale, kqmax_scale);
+#pragma unroll
+        for (int i_VKQ_0 = 0; i_VKQ_0 < D/2; i_VKQ_0 += nthreads_V) {
+            VKQ[j_VKQ][i_VKQ_0/nthreads_V] *= kqmax_scale_h2;
+        }
+#pragma unroll
+        for (int i_VKQ_0 = 0; i_VKQ_0 < D/2; i_VKQ_0 += nthreads_V*V_rows_per_thread/2) {
+            const int i_VKQ = i_VKQ_0 + (nthreads_V == WARP_SIZE ? threadIdx.x : threadIdx.x % nthreads_V)*(V_rows_per_thread/2);
+
+            ggml_cuda_memcpy_1<V_rows_per_thread*sizeof(half)>(VKQ_tmp + i_VKQ, &VKQ[j_VKQ][i_VKQ_0/nthreads_V]);
+        }
+#else
+        float2 * VKQ_tmp = (float2 *) KQ + threadIdx.y*(V_cols_per_iter*D/2)
+            + (nthreads_V == WARP_SIZE ? 0 : threadIdx.x / nthreads_V)*(D/2);
+
+#pragma unroll
+        for (int i_VKQ_0 = 0; i_VKQ_0 < D/2; i_VKQ_0 += nthreads_V) {
+            VKQ[j_VKQ][i_VKQ_0/nthreads_V].x *= kqmax_scale;
+            VKQ[j_VKQ][i_VKQ_0/nthreads_V].y *= kqmax_scale;
+        }
+#pragma unroll
+        for (int i_VKQ_0 = 0; i_VKQ_0 < D/2; i_VKQ_0 += nthreads_V*V_rows_per_thread/2) {
+            const int i_VKQ = i_VKQ_0 + (nthreads_V == WARP_SIZE ? threadIdx.x : threadIdx.x % nthreads_V)*(V_rows_per_thread/2);
+
+            ggml_cuda_memcpy_1<V_rows_per_thread/2*sizeof(float)>(VKQ_tmp + i_VKQ,                       &VKQ[j_VKQ][i_VKQ_0/nthreads_V]);
+            ggml_cuda_memcpy_1<V_rows_per_thread/2*sizeof(float)>(VKQ_tmp + i_VKQ + V_rows_per_thread/4, &VKQ[j_VKQ][i_VKQ_0/nthreads_V + V_rows_per_thread/4]);
+        }
+#endif // FAST_FP16_AVAILABLE
+
+        KQ_sum[j_VKQ] *= kqmax_scale;
+        KQ_sum[j_VKQ] = warp_reduce_sum(KQ_sum[j_VKQ]);
+        if (threadIdx.x == 0) {
+            KQ_sum_shared[j_VKQ][threadIdx.y] = KQ_sum[j_VKQ];
+        }
+
+        __syncthreads();
+
+        if (nthreads <= D || tid < D) {
+            KQ_sum[j_VKQ] = KQ_sum_shared[j_VKQ][threadIdx.x];
+            KQ_sum[j_VKQ] = warp_reduce_sum(KQ_sum[j_VKQ]);
+
+#pragma unroll
+            for (int i0 = 0; i0 < D; i0 += nthreads) {
+                float dst_val = 0;
+#pragma unroll
+                for (int w = 0; w < nwarps; ++w) {
+#pragma unroll
+                    for (int v = 0; v < V_cols_per_iter; ++v) {
+                        dst_val += float(KQ[w*V_cols_per_iter*D + v*D + i0 + tid]);
+                    }
+                }
+                if (gridDim.y == 1) {
+                    dst_val /= KQ_sum[j_VKQ];
+                }
+                dst[(((sequence*ne01 + ic0 + j_VKQ)*ne02 + head)*gridDim.y + blockIdx.y)*D + i0 + tid] = dst_val;
+            }
+        }
+
+        if (j_VKQ < ncols-1) {
+            __syncthreads();
+        }
+
+    }
+
+    if (gridDim.y != 1 && tid < ncols && (ncols == 1 || ic0 + tid < ne01)) {
+        dst_meta[((sequence*ne01 + ic0 + tid)*ne02 + head)*gridDim.y + blockIdx.y] = make_float2(KQ_max[tid], KQ_sum[tid]);
+    }
+#else
+    GGML_UNUSED_VARS(Q, K, V, mask, sinks, KV_max, dst, dst_meta, scale,
+        max_bias, m0, m1, n_head_log2, logit_softcap,
+        ne00, ne01, ne02, ne03,
+              nb01, nb02, nb03,
+        ne10, ne11, ne12, ne13,
+              nb11, nb12, nb13,
+              nb21, nb22, nb23,
+              ne31, ne32, ne33,
+              nb31, nb32, nb33);
+    NO_DEVICE_CODE;
+#endif // FLASH_ATTN_AVAILABLE
+}
+#ifdef __clang__
+#pragma clang diagnostic pop
+#endif // __clang__
+
+template <int D, int cols_per_block, ggml_type type_K, ggml_type type_V, bool use_logit_softcap>
+void ggml_cuda_flash_attn_ext_vec_case_impl(ggml_backend_cuda_context & ctx, ggml_tensor * dst) {
+    const int cc = ggml_cuda_info().devices[ggml_cuda_get_device()].cc;
+
+    const int nthreads = ggml_cuda_fattn_vec_get_nthreads_host(cc);
+    const int nwarps   = nthreads / WARP_SIZE;
+    fattn_kernel_t fattn_kernel = flash_attn_ext_vec<D, cols_per_block, type_K, type_V, use_logit_softcap>;
+    constexpr bool need_f16_K = false;
+    constexpr bool need_f16_V = false;
+    constexpr size_t nbytes_shared = 0;
+    launch_fattn<D, cols_per_block, 1>(ctx, dst, fattn_kernel, nwarps, nbytes_shared, D, need_f16_K, need_f16_V, false);
+}
+
+template <int D, ggml_type type_K, ggml_type type_V>
+void ggml_cuda_flash_attn_ext_vec_case(ggml_backend_cuda_context & ctx, ggml_tensor * dst) {
+    const ggml_tensor * KQV = dst;
+    const ggml_tensor * Q   = dst->src[0];
+    const ggml_tensor * K   = dst->src[1];
+    const ggml_tensor * V   = dst->src[2];
+
+    GGML_ASSERT(K->type == type_K);
+    GGML_ASSERT(V->type == type_V);
+
+    float logit_softcap;
+    memcpy(&logit_softcap, (const float *) KQV->op_params + 2, sizeof(float));
+
+    const int cc = ggml_cuda_info().devices[ggml_cuda_get_device()].cc;
+
+    if (Q->ne[1] == 1) {
+        constexpr int cols_per_block = 1;
+        if (logit_softcap == 0.0f) {
+            constexpr bool use_logit_softcap = false;
+            ggml_cuda_flash_attn_ext_vec_case_impl<D, cols_per_block, type_K, type_V, use_logit_softcap>(ctx, dst);
+        } else {
+            constexpr bool use_logit_softcap = true;
+            ggml_cuda_flash_attn_ext_vec_case_impl<D, cols_per_block, type_K, type_V, use_logit_softcap>(ctx, dst);
+        }
+        return;
+    }
+
+    constexpr int cols_per_block = 2;
+    if (logit_softcap == 0.0f) {
+        constexpr bool use_logit_softcap = false;
+        ggml_cuda_flash_attn_ext_vec_case_impl<D, cols_per_block, type_K, type_V, use_logit_softcap>(ctx, dst);
+    } else {
+        constexpr bool use_logit_softcap = true;
+        ggml_cuda_flash_attn_ext_vec_case_impl<D, cols_per_block, type_K, type_V, use_logit_softcap>(ctx, dst);
+    }
+}
+
+#define DECL_FATTN_VEC_CASE(D, type_K, type_V)                              \
+    template void ggml_cuda_flash_attn_ext_vec_case                         \
+    <D, type_K, type_V>(ggml_backend_cuda_context & ctx, ggml_tensor * dst) \
+
+#define EXTERN_DECL_FATTN_VEC_CASES(D, type_K)             \
+    extern DECL_FATTN_VEC_CASE(D, type_K, GGML_TYPE_F16);  \
+    extern DECL_FATTN_VEC_CASE(D, type_K, GGML_TYPE_Q4_0); \
+    extern DECL_FATTN_VEC_CASE(D, type_K, GGML_TYPE_Q4_1); \
+    extern DECL_FATTN_VEC_CASE(D, type_K, GGML_TYPE_Q5_0); \
+    extern DECL_FATTN_VEC_CASE(D, type_K, GGML_TYPE_Q5_1); \
+    extern DECL_FATTN_VEC_CASE(D, type_K, GGML_TYPE_Q8_0); \
+
+EXTERN_DECL_FATTN_VEC_CASES( 64, GGML_TYPE_F16)
+EXTERN_DECL_FATTN_VEC_CASES( 64, GGML_TYPE_Q4_0)
+EXTERN_DECL_FATTN_VEC_CASES( 64, GGML_TYPE_Q4_1)
+EXTERN_DECL_FATTN_VEC_CASES( 64, GGML_TYPE_Q5_0)
+EXTERN_DECL_FATTN_VEC_CASES( 64, GGML_TYPE_Q5_1)
+EXTERN_DECL_FATTN_VEC_CASES( 64, GGML_TYPE_Q8_0)
+
+EXTERN_DECL_FATTN_VEC_CASES(128, GGML_TYPE_F16)
+EXTERN_DECL_FATTN_VEC_CASES(128, GGML_TYPE_Q4_0)
+EXTERN_DECL_FATTN_VEC_CASES(128, GGML_TYPE_Q4_1)
+EXTERN_DECL_FATTN_VEC_CASES(128, GGML_TYPE_Q5_0)
+EXTERN_DECL_FATTN_VEC_CASES(128, GGML_TYPE_Q5_1)
+EXTERN_DECL_FATTN_VEC_CASES(128, GGML_TYPE_Q8_0)
+
+EXTERN_DECL_FATTN_VEC_CASES(256, GGML_TYPE_F16)
+EXTERN_DECL_FATTN_VEC_CASES(256, GGML_TYPE_Q4_0)
+EXTERN_DECL_FATTN_VEC_CASES(256, GGML_TYPE_Q4_1)
+EXTERN_DECL_FATTN_VEC_CASES(256, GGML_TYPE_Q5_0)
+EXTERN_DECL_FATTN_VEC_CASES(256, GGML_TYPE_Q5_1)
+EXTERN_DECL_FATTN_VEC_CASES(256, GGML_TYPE_Q8_0)