CUDA: faster large batch FA without tensor cores (llama/7314)

ggml-cuda/fattn-tile-f16.cu

@@ -0,0 +1,395 @@
+#include "common.cuh"
+#include "fattn-common.cuh"
+#include "fattn-tile-f16.cuh"
+
+#define FATTN_KQ_STRIDE_TILE_F16 64
+
+template<int D, int ncols, int nwarps, int parallel_blocks> // D == head size
+#if !(defined(GGML_USE_HIPBLAS) && defined(__HIP_PLATFORM_AMD__))
+__launch_bounds__(nwarps*WARP_SIZE, 1)
+#endif // !(defined(GGML_USE_HIPBLAS) && defined(__HIP_PLATFORM_AMD__))
+static __global__ void flash_attn_tile_ext_f16(
+        const char * __restrict__ Q,
+        const char * __restrict__ K,
+        const char * __restrict__ V,
+        const char * __restrict__ mask,
+        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 int ne00,
+        const int ne01,
+        const int ne02,
+        const int ne03,
+        const int ne10,
+        const int ne11,
+        const int ne12,
+        const int ne13,
+        const int ne31,
+        const int nb31,
+        const int nb01,
+        const int nb02,
+        const int nb03,
+        const int nb11,
+        const int nb12,
+        const int nb13,
+        const int ne0,
+        const int ne1,
+        const int ne2,
+        const int ne3) {
+#if FP16_AVAILABLE
+    //In this kernel Q, K, V are matrices while i, j, k are matrix indices.
+
+    const int ic0 = (blockIdx.x / parallel_blocks) * ncols; // Index of the Q/QKV column to work on.
+    const int ip  =  blockIdx.x % parallel_blocks; // Index in group of blocks running for the same column in parallel.
+
+    const int gqa_ratio = ne02 / ne12; // With grouped query attention there are > 1 Q matrices per K, V matrix.
+    const float2 * Q_f2  = (const float2 *) (Q    + nb02* blockIdx.y              + nb01*ic0);
+    const half2  * K_h2  = (const half2  *) (K    + nb12*(blockIdx.y / gqa_ratio));
+    const half2  * V_h2  = (const half2  *) (V    + nb12*(blockIdx.y / gqa_ratio)); // K and V have same shape
+    const half   * maskh = (const half   *)  mask + ne11*ic0;
+
+    const int stride_KV2 = nb11 / sizeof(half2);
+
+    half slopeh = __float2half(1.0f);
+
+    // ALiBi
+    if (max_bias > 0.0f) {
+        const uint32_t h = blockIdx.y;
+
+        const float base = h < n_head_log2 ? m0 : m1;
+        const int   exph = h < n_head_log2 ? h + 1 : 2*(h - n_head_log2) + 1;
+
+        slopeh = __float2half(powf(base, exph));
+    }
+
+    static_assert(D % (2*WARP_SIZE) == 0, "D not divisible by 2*WARP_SIZE == 64.");
+
+    __shared__ half KQ[ncols*FATTN_KQ_STRIDE_TILE_F16];
+    half2 * KQ2 = (half2 *) KQ;
+
+    __shared__ half2 KV_tmp[FATTN_KQ_STRIDE_TILE_F16][D/2 + 1]; // Pad D to avoid memory bank conflicts.
+
+    half kqmax[ncols/nwarps];
+#pragma unroll
+    for (int j0 = 0; j0 < ncols; j0 += nwarps) {
+        kqmax[j0/nwarps] = -HALF_MAX_HALF;
+    }
+    half2 kqsum[ncols/nwarps] = {{0.0f, 0.0f}};
+
+    half2 VKQ[ncols/nwarps][(D/2)/WARP_SIZE] = {{{0.0f, 0.0f}}};
+
+    // Convert Q to half2 and store in registers:
+    __shared__ half2 Q_h2[ncols][D/2];
+#pragma unroll
+    for (int j0 = 0; j0 < ncols; j0 += nwarps) {
+        const int j = j0 + threadIdx.y;
+
+#pragma unroll
+        for (int i0 = 0; i0 < D/2; i0 += WARP_SIZE) {
+            const int i = i0 + threadIdx.x;
+
+            const float2 tmp = Q_f2[j*(nb01/sizeof(float2)) + i];
+            Q_h2[j][i] = make_half2(scale, scale) * make_half2(tmp.x, tmp.y);
+        }
+    }
+
+    __syncthreads();
+
+    const int k_start = parallel_blocks == 1 ? 0 : ip*FATTN_KQ_STRIDE_TILE_F16;
+    for (int k_VKQ_0 = k_start; k_VKQ_0 < ne11; k_VKQ_0 += parallel_blocks*FATTN_KQ_STRIDE_TILE_F16) {
+        // Calculate KQ tile and keep track of new maximum KQ values:
+
+        half kqmax_new[ncols/nwarps];
+#pragma unroll
+        for (int j = 0; j < ncols/nwarps; ++j) {
+            kqmax_new[j] = kqmax[j];
+        }
+
+#pragma unroll
+        for (int i_KQ_0 = 0; i_KQ_0 < FATTN_KQ_STRIDE_TILE_F16; i_KQ_0 += nwarps) {
+            const int i_KQ = i_KQ_0 + threadIdx.y;
+
+#pragma unroll
+            for (int k_KQ_0 = 0; k_KQ_0 < D/2; k_KQ_0 += WARP_SIZE) {
+                const int k_KQ = k_KQ_0 + threadIdx.x;
+
+                KV_tmp[i_KQ][k_KQ] = K_h2[(k_VKQ_0 + i_KQ)*stride_KV2 + k_KQ];
+            }
+        }
+
+        __syncthreads();
+
+        half2 sum2[FATTN_KQ_STRIDE_TILE_F16/WARP_SIZE][ncols/nwarps] = {{{0.0f, 0.0f}}};
+
+#pragma unroll
+        for (int k_KQ = 0; k_KQ < D/2; ++k_KQ) {
+            half2 K_k[FATTN_KQ_STRIDE_TILE_F16/WARP_SIZE];
+            half2 Q_k[ncols/nwarps];
+
+#pragma unroll
+            for (int i_KQ_0 = 0; i_KQ_0 < FATTN_KQ_STRIDE_TILE_F16; i_KQ_0 += WARP_SIZE) {
+                const int i_KQ = i_KQ_0 + threadIdx.x;
+
+                K_k[i_KQ_0/WARP_SIZE] = KV_tmp[i_KQ][k_KQ];
+            }
+#pragma unroll
+            for (int j_KQ_0 = 0; j_KQ_0 < ncols; j_KQ_0 += nwarps) {
+                const int j_KQ = j_KQ_0 + threadIdx.y;
+
+                Q_k[j_KQ_0/nwarps] = Q_h2[j_KQ][k_KQ];
+            }
+
+#pragma unroll
+            for (int i_KQ_0 = 0; i_KQ_0 < FATTN_KQ_STRIDE_TILE_F16; i_KQ_0 += WARP_SIZE) {
+#pragma unroll
+                for (int j_KQ_0 = 0; j_KQ_0 < ncols; j_KQ_0 += nwarps) {
+                    sum2[i_KQ_0/WARP_SIZE][j_KQ_0/nwarps] += K_k[i_KQ_0/WARP_SIZE]*Q_k[j_KQ_0/nwarps];
+                }
+            }
+        }
+
+#pragma unroll
+        for (int i_KQ_0 = 0; i_KQ_0 < FATTN_KQ_STRIDE_TILE_F16; i_KQ_0 += WARP_SIZE) {
+            const int i_KQ = i_KQ_0 + threadIdx.x;
+
+#pragma unroll
+            for (int j_KQ_0 = 0; j_KQ_0 < ncols; j_KQ_0 += nwarps) {
+                const int j_KQ = j_KQ_0 + threadIdx.y;
+
+                half sum = __low2half(sum2[i_KQ_0/WARP_SIZE][j_KQ_0/nwarps]) + __high2half(sum2[i_KQ_0/WARP_SIZE][j_KQ_0/nwarps]);
+                sum += mask ? slopeh*maskh[j_KQ*ne11 + k_VKQ_0 + i_KQ] : __float2half(0.0f);
+
+                kqmax_new[j_KQ_0/nwarps] = ggml_cuda_hmax(kqmax_new[j_KQ_0/nwarps], sum);
+
+                KQ[j_KQ*FATTN_KQ_STRIDE_TILE_F16 + i_KQ] = sum;
+            }
+        }
+
+        __syncthreads();
+
+#pragma unroll
+        for (int j0 = 0; j0 < ncols; j0 += nwarps) {
+            const int j = j0 + threadIdx.y;
+
+            kqmax_new[j0/nwarps] = warp_reduce_max(kqmax_new[j0/nwarps]);
+            const half2 KQ_max_scale = __half2half2(hexp(kqmax[j0/nwarps] - kqmax_new[j0/nwarps]));
+            kqmax[j0/nwarps] = kqmax_new[j0/nwarps];
+
+#pragma unroll
+            for (int i0 = 0; i0 < FATTN_KQ_STRIDE_TILE_F16/2; i0 += WARP_SIZE) {
+                const int i = i0 + threadIdx.x;
+
+                const half2 diff = KQ2[j*(FATTN_KQ_STRIDE_TILE_F16/2) + i] - __half2half2(kqmax[j0/nwarps]);
+                const half2 val = h2exp(diff);
+                kqsum[j0/nwarps] = kqsum[j0/nwarps]*KQ_max_scale + val;
+                KQ2[j*(FATTN_KQ_STRIDE_TILE_F16/2) + i] = val;
+            }
+
+#pragma unroll
+            for (int i0 = 0; i0 < D/2; i0 += WARP_SIZE) {
+                VKQ[j0/nwarps][i0/WARP_SIZE] *= KQ_max_scale;
+            }
+        }
+
+        __syncthreads();
+
+#pragma unroll
+        for (int k0 = 0; k0 < FATTN_KQ_STRIDE_TILE_F16; k0 += nwarps) {
+            const int k = k0 + threadIdx.y;
+
+#pragma unroll
+            for (int i0 = 0; i0 < D/2; i0 += WARP_SIZE) {
+                const int i = i0 + threadIdx.x;
+
+                KV_tmp[k][i] = V_h2[(k_VKQ_0 + k)*stride_KV2 + i];
+            }
+        }
+
+        __syncthreads();
+
+#pragma unroll
+        for (int k0 = 0; k0 < FATTN_KQ_STRIDE_TILE_F16; k0 += 2) {
+            half2  V_k[(D/2)/WARP_SIZE][2];
+            half2 KQ_k[ncols/nwarps];
+
+#pragma unroll
+            for (int i0 = 0; i0 < D/2; i0 += WARP_SIZE) {
+                const int i = i0 + threadIdx.x;
+
+                V_k[i0/WARP_SIZE][0] = KV_tmp[k0 + 0][i];
+                V_k[i0/WARP_SIZE][1] = KV_tmp[k0 + 1][i];
+            }
+#pragma unroll
+            for (int j0 = 0; j0 < ncols; j0 += nwarps) {
+                const int j = j0 + threadIdx.y;
+
+                KQ_k[j0/nwarps] = KQ2[j*(FATTN_KQ_STRIDE_TILE_F16/2) + k0/2];
+            }
+
+#pragma unroll
+            for (int i0 = 0; i0 < D/2; i0 += WARP_SIZE) {
+#pragma unroll
+                for (int j0 = 0; j0 < ncols; j0 += nwarps) {
+                    VKQ[j0/nwarps][i0/WARP_SIZE] += V_k[i0/WARP_SIZE][0]* __low2half2(KQ_k[j0/nwarps]);
+                    VKQ[j0/nwarps][i0/WARP_SIZE] += V_k[i0/WARP_SIZE][1]*__high2half2(KQ_k[j0/nwarps]);
+                }
+            }
+        }
+
+        __syncthreads();
+    }
+
+#pragma unroll
+    for (int j_VKQ_0 = 0; j_VKQ_0 < ncols; j_VKQ_0 += nwarps) {
+        const int j_VKQ = j_VKQ_0 + threadIdx.y;
+
+        half kqsum_j = __low2half(kqsum[j_VKQ_0/nwarps]) + __high2half(kqsum[j_VKQ_0/nwarps]);
+        kqsum_j = warp_reduce_sum(kqsum_j);
+
+#pragma unroll
+        for (int i00 = 0; i00 < D; i00 += 2*WARP_SIZE) {
+            const int i0 = i00 + 2*threadIdx.x;
+
+            half2 dst_val = VKQ[j_VKQ_0/nwarps][i0/(2*WARP_SIZE)];
+            if (parallel_blocks == 1) {
+                dst_val /= __half2half2(kqsum_j);
+            }
+            const int j_dst = (ic0 + j_VKQ)*parallel_blocks + ip;
+            dst[j_dst*D*gridDim.y + D*blockIdx.y + i0 + 0] =  __low2float(dst_val);
+            dst[j_dst*D*gridDim.y + D*blockIdx.y + i0 + 1] = __high2float(dst_val);
+        }
+
+        if (parallel_blocks != 1 && threadIdx.x == 0) {
+            dst_meta[(ic0 + j_VKQ)*gridDim.y*parallel_blocks + blockIdx.y*parallel_blocks + ip] = make_float2(kqmax[j_VKQ_0/nwarps], kqsum_j);
+        }
+    }
+#else
+   NO_DEVICE_CODE;
+#endif // FP16_AVAILABLE
+}
+
+template <int D, int cols_per_block, int parallel_blocks> void launch_fattn_tile_f16(
+        const ggml_tensor * Q, const ggml_tensor * K, const ggml_tensor * V, ggml_tensor * KQV, const ggml_tensor * mask,
+        ggml_cuda_pool & pool, cudaStream_t main_stream
+) {
+    ggml_cuda_pool_alloc<float>  dst_tmp(pool);
+    ggml_cuda_pool_alloc<float2> dst_tmp_meta(pool);
+
+    if (parallel_blocks > 1) {
+        dst_tmp.alloc(parallel_blocks*ggml_nelements(KQV));
+        dst_tmp_meta.alloc(parallel_blocks*ggml_nrows(KQV));
+    }
+
+    constexpr int  nwarps = 8;
+    const     dim3 block_dim(WARP_SIZE, nwarps, 1);
+    const     dim3 blocks_num(parallel_blocks*((Q->ne[1] + cols_per_block - 1) / cols_per_block), Q->ne[2], Q->ne[3]);
+    const     int  shmem = 0;
+
+    float scale    = 1.0f;
+    float max_bias = 0.0f;
+
+    memcpy(&scale,    (float *) KQV->op_params + 0, sizeof(float));
+    memcpy(&max_bias, (float *) KQV->op_params + 1, sizeof(float));
+
+    const uint32_t n_head      = Q->ne[2];
+    const uint32_t n_head_log2 = 1u << (uint32_t) floorf(log2f((float) n_head));
+
+    const float m0 = powf(2.0f, -(max_bias       ) / n_head_log2);
+    const float m1 = powf(2.0f, -(max_bias / 2.0f) / n_head_log2);
+
+    flash_attn_tile_ext_f16<D, cols_per_block, nwarps, parallel_blocks>
+        <<<blocks_num, block_dim, shmem, main_stream>>> (
+                (const char *) Q->data,
+                (const char *) K->data,
+                (const char *) V->data,
+                mask ? ((const char *) mask->data) : nullptr,
+                parallel_blocks == 1 ? (float *) KQV->data : dst_tmp.ptr, dst_tmp_meta.ptr,
+                scale, max_bias, m0, m1, n_head_log2,
+                Q->ne[0], Q->ne[1], Q->ne[2], Q->ne[3],
+                K->ne[0], K->ne[1], K->ne[2], K->ne[3],
+                mask ? mask->ne[1] : 0, mask ?  mask->nb[1] : 0,
+                Q->nb[1], Q->nb[2], Q->nb[3],
+                K->nb[1], K->nb[2], K->nb[3],
+                KQV->ne[0], KQV->ne[1], KQV->ne[2], KQV->ne[3]
+                );
+    CUDA_CHECK(cudaGetLastError());
+
+    if (parallel_blocks == 1) {
+        return;
+    }
+
+    const dim3 block_dim_combine(D, 1, 1);
+    const dim3 blocks_num_combine(Q->ne[1], blocks_num.y, blocks_num.z);
+    const int  shmem_combine = 0;
+
+    flash_attn_combine_results<D, parallel_blocks>
+        <<<blocks_num_combine, block_dim_combine, shmem_combine, main_stream>>>
+        (dst_tmp.ptr, dst_tmp_meta.ptr, (float *) KQV->data);
+    CUDA_CHECK(cudaGetLastError());
+}
+
+void ggml_cuda_flash_attn_ext_tile_f16(ggml_backend_cuda_context & ctx, ggml_tensor * dst) {
+    const ggml_tensor * Q = dst->src[0];
+    const ggml_tensor * K = dst->src[1];
+    const ggml_tensor * V = dst->src[2];
+
+    const ggml_tensor * mask = dst->src[3];
+
+    ggml_tensor * KQV = dst;
+
+    const int32_t precision = KQV->op_params[2];
+    GGML_ASSERT(precision == GGML_PREC_DEFAULT);
+    GGML_ASSERT(Q->ne[0] == 64 || Q->ne[0] == 128 && "FlashAttention without tensor cores only supports head sizes 64 and 128.");
+
+    if (Q->ne[1] <= 16) {
+        constexpr int cols_per_block = 16;
+        constexpr int parallel_blocks = 4;
+        switch (Q->ne[0]) {
+            case 64:
+                launch_fattn_tile_f16< 64, cols_per_block, parallel_blocks>(Q, K, V, KQV, mask, ctx.pool(), ctx.stream());
+                break;
+            case 128:
+                launch_fattn_tile_f16<128, cols_per_block, parallel_blocks>(Q, K, V, KQV, mask, ctx.pool(), ctx.stream());
+                break;
+            default:
+                GGML_ASSERT(false);
+                break;
+        }
+        return;
+    }
+
+    if (Q->ne[1] <= 32) {
+        constexpr int cols_per_block = 32;
+        constexpr int parallel_blocks = 4;
+        switch (Q->ne[0]) {
+            case 64:
+                launch_fattn_tile_f16< 64, cols_per_block, parallel_blocks>(Q, K, V, KQV, mask, ctx.pool(), ctx.stream());
+                break;
+            case 128:
+                launch_fattn_tile_f16<128, cols_per_block, parallel_blocks>(Q, K, V, KQV, mask, ctx.pool(), ctx.stream());
+                break;
+            default:
+                GGML_ASSERT(false);
+                break;
+        }
+        return;
+    }
+
+    constexpr int cols_per_block = 32;
+    constexpr int parallel_blocks = 1;
+    switch (Q->ne[0]) {
+        case 64:
+            launch_fattn_tile_f16< 64, cols_per_block, parallel_blocks>(Q, K, V, KQV, mask, ctx.pool(), ctx.stream());
+            break;
+        case 128:
+            launch_fattn_tile_f16<128, cols_per_block, parallel_blocks>(Q, K, V, KQV, mask, ctx.pool(), ctx.stream());
+            break;
+        default:
+            GGML_ASSERT(false);
+            break;
+    }
+}

ggml-cuda/fattn-tile-f16.cuh

@@ -0,0 +1,3 @@
+#include "common.cuh"
+
+void ggml_cuda_flash_attn_ext_tile_f16(ggml_backend_cuda_context & ctx, ggml_tensor * dst);

ggml-cuda/fattn-tile-f32.cu

@@ -0,0 +1,393 @@
+#include "common.cuh"
+#include "fattn-common.cuh"
+#include "fattn-tile-f32.cuh"
+
+#define FATTN_KQ_STRIDE_TILE_F32 32
+
+template<int D, int ncols, int nwarps, int parallel_blocks> // D == head size
+#if !(defined(GGML_USE_HIPBLAS) && defined(__HIP_PLATFORM_AMD__))
+__launch_bounds__(nwarps*WARP_SIZE, 1)
+#endif // !(defined(GGML_USE_HIPBLAS) && defined(__HIP_PLATFORM_AMD__))
+static __global__ void flash_attn_tile_ext_f32(
+        const char * __restrict__ Q,
+        const char * __restrict__ K,
+        const char * __restrict__ V,
+        const char * __restrict__ mask,
+        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 int ne00,
+        const int ne01,
+        const int ne02,
+        const int ne03,
+        const int ne10,
+        const int ne11,
+        const int ne12,
+        const int ne13,
+        const int ne31,
+        const int nb31,
+        const int nb01,
+        const int nb02,
+        const int nb03,
+        const int nb11,
+        const int nb12,
+        const int nb13,
+        const int ne0,
+        const int ne1,
+        const int ne2,
+        const int ne3) {
+    //In this kernel Q, K, V are matrices while i, j, k are matrix indices.
+
+    const int ic0 = (blockIdx.x / parallel_blocks) * ncols; // Index of the Q/QKV column to work on.
+    const int ip  =  blockIdx.x % parallel_blocks; // Index in group of blocks running for the same column in parallel.
+
+    const int gqa_ratio = ne02 / ne12; // With grouped query attention there are > 1 Q matrices per K, V matrix.
+    const float2 * Q_f2  = (const float2 *) (Q    + nb02* blockIdx.y              + nb01*ic0);
+    const half2  * K_h2  = (const half2  *) (K    + nb12*(blockIdx.y / gqa_ratio));
+    const half2  * V_h2  = (const half2  *) (V    + nb12*(blockIdx.y / gqa_ratio)); // K and V have same shape
+    const half   * maskh = (const half   *)  mask + ne11*ic0;
+
+    const int stride_KV2 = nb11 / sizeof(half2);
+
+    float slope = 1.0f;
+
+    // ALiBi
+    if (max_bias > 0.0f) {
+        const uint32_t h = blockIdx.y;
+
+        const float base = h < n_head_log2 ? m0 : m1;
+        const int   exph = h < n_head_log2 ? h + 1 : 2*(h - n_head_log2) + 1;
+
+        slope = powf(base, exph);
+    }
+
+    static_assert(D % (2*WARP_SIZE) == 0, "D not divisible by 2*WARP_SIZE == 64.");
+
+    __shared__ float KQ[ncols*FATTN_KQ_STRIDE_TILE_F32];
+
+    __shared__ float KV_tmp[FATTN_KQ_STRIDE_TILE_F32][D + 1]; // Pad D to avoid memory bank conflicts.
+    float2 * KV_tmp2 = (float2 *) KV_tmp;
+
+    float kqmax[ncols/nwarps];
+#pragma unroll
+    for (int j0 = 0; j0 < ncols; j0 += nwarps) {
+        kqmax[j0/nwarps] = -FLT_MAX/2.0f;
+    }
+    float kqsum[ncols/nwarps] = {0.0f};
+
+    float2 VKQ[ncols/nwarps][(D/2)/WARP_SIZE] = {{{0.0f, 0.0f}}};
+
+    // Convert Q to half2 and store in registers:
+    __shared__ float Q_f[ncols][D];
+#pragma unroll
+    for (int j0 = 0; j0 < ncols; j0 += nwarps) {
+        const int j = j0 + threadIdx.y;
+
+#pragma unroll
+        for (int i0 = 0; i0 < D; i0 += 2*WARP_SIZE) {
+            float2 tmp = Q_f2[j*(nb01/sizeof(float2)) + i0/2 + threadIdx.x];
+            Q_f[j][i0 + 0*WARP_SIZE + threadIdx.x] = tmp.x * scale;
+            Q_f[j][i0 + 1*WARP_SIZE + threadIdx.x] = tmp.y * scale;
+        }
+    }
+
+    __syncthreads();
+
+    const int k_start = parallel_blocks == 1 ? 0 : ip*FATTN_KQ_STRIDE_TILE_F32;
+    for (int k_VKQ_0 = k_start; k_VKQ_0 < ne11; k_VKQ_0 += parallel_blocks*FATTN_KQ_STRIDE_TILE_F32) {
+        // Calculate KQ tile and keep track of new maximum KQ values:
+
+        float kqmax_new[ncols/nwarps];
+#pragma unroll
+        for (int j = 0; j < ncols/nwarps; ++j) {
+            kqmax_new[j] = kqmax[j];
+        }
+
+#pragma unroll
+        for (int i_KQ_0 = 0; i_KQ_0 < FATTN_KQ_STRIDE_TILE_F32; i_KQ_0 += nwarps) {
+            const int i_KQ = i_KQ_0 + threadIdx.y;
+
+#pragma unroll
+            for (int k_KQ_0 = 0; k_KQ_0 < D; k_KQ_0 += 2*WARP_SIZE) {
+                const half2 tmp = K_h2[(k_VKQ_0 + i_KQ)*stride_KV2 + k_KQ_0/2 + threadIdx.x];
+                KV_tmp[i_KQ][k_KQ_0 + 0*WARP_SIZE + threadIdx.x] =  __low2float(tmp);
+                KV_tmp[i_KQ][k_KQ_0 + 1*WARP_SIZE + threadIdx.x] = __high2float(tmp);
+            }
+        }
+
+        __syncthreads();
+
+        float sum[FATTN_KQ_STRIDE_TILE_F32/WARP_SIZE][ncols/nwarps] = {{0.0f}};
+
+#pragma unroll
+        for (int k_KQ = 0; k_KQ < D; ++k_KQ) {
+            float K_k[FATTN_KQ_STRIDE_TILE_F32/WARP_SIZE];
+            float Q_k[ncols/nwarps];
+
+#pragma unroll
+            for (int i_KQ_0 = 0; i_KQ_0 < FATTN_KQ_STRIDE_TILE_F32; i_KQ_0 += WARP_SIZE) {
+                const int i_KQ = i_KQ_0 + threadIdx.x;
+
+                K_k[i_KQ_0/WARP_SIZE] = KV_tmp[i_KQ][k_KQ];
+            }
+#pragma unroll
+            for (int j_KQ_0 = 0; j_KQ_0 < ncols; j_KQ_0 += nwarps) {
+                const int j_KQ = j_KQ_0 + threadIdx.y;
+
+                Q_k[j_KQ_0/nwarps] = Q_f[j_KQ][k_KQ];
+            }
+
+#pragma unroll
+            for (int i_KQ_0 = 0; i_KQ_0 < FATTN_KQ_STRIDE_TILE_F32; i_KQ_0 += WARP_SIZE) {
+#pragma unroll
+                for (int j_KQ_0 = 0; j_KQ_0 < ncols; j_KQ_0 += nwarps) {
+                    sum[i_KQ_0/WARP_SIZE][j_KQ_0/nwarps] += K_k[i_KQ_0/WARP_SIZE] * Q_k[j_KQ_0/nwarps];
+                }
+            }
+        }
+
+#pragma unroll
+        for (int i_KQ_0 = 0; i_KQ_0 < FATTN_KQ_STRIDE_TILE_F32; i_KQ_0 += WARP_SIZE) {
+            const int i_KQ = i_KQ_0 + threadIdx.x;
+
+#pragma unroll
+            for (int j_KQ_0 = 0; j_KQ_0 < ncols; j_KQ_0 += nwarps) {
+                const int j_KQ = j_KQ_0 + threadIdx.y;
+
+                sum[i_KQ_0/WARP_SIZE][j_KQ_0/nwarps] += mask ? slope*__half2float(maskh[j_KQ*ne11 + k_VKQ_0 + i_KQ]) : 0.0f;
+
+                kqmax_new[j_KQ_0/nwarps] = fmaxf(kqmax_new[j_KQ_0/nwarps], sum[i_KQ_0/WARP_SIZE][j_KQ_0/nwarps]);
+
+                KQ[j_KQ*FATTN_KQ_STRIDE_TILE_F32 + i_KQ] = sum[i_KQ_0/WARP_SIZE][j_KQ_0/nwarps];
+            }
+        }
+
+        __syncthreads();
+
+#pragma unroll
+        for (int j0 = 0; j0 < ncols; j0 += nwarps) {
+            const int j = j0 + threadIdx.y;
+
+            kqmax_new[j0/nwarps] = warp_reduce_max(kqmax_new[j0/nwarps]);
+            const float KQ_max_scale = expf(kqmax[j0/nwarps] - kqmax_new[j0/nwarps]);
+            kqmax[j0/nwarps] = kqmax_new[j0/nwarps];
+
+            float kqsum_add = 0.0f;
+#pragma unroll
+            for (int i0 = 0; i0 < FATTN_KQ_STRIDE_TILE_F32; i0 += WARP_SIZE) {
+                const int i = i0 + threadIdx.x;
+
+                const float diff = KQ[j*FATTN_KQ_STRIDE_TILE_F32 + i] - kqmax[j0/nwarps];
+                const float val = expf(diff);
+                kqsum_add += val;
+                KQ[j*FATTN_KQ_STRIDE_TILE_F32 + i] = val;
+            }
+            kqsum[j0/nwarps] = kqsum[j0/nwarps]*KQ_max_scale + kqsum_add;
+
+#pragma unroll
+            for (int i0 = 0; i0 < D/2; i0 += WARP_SIZE) {
+                VKQ[j0/nwarps][i0/WARP_SIZE].x *= KQ_max_scale;
+                VKQ[j0/nwarps][i0/WARP_SIZE].y *= KQ_max_scale;
+            }
+        }
+
+        __syncthreads();
+
+#pragma unroll
+        for (int k0 = 0; k0 < FATTN_KQ_STRIDE_TILE_F32; k0 += nwarps) {
+            const int k = k0 + threadIdx.y;
+
+#pragma unroll
+            for (int i0 = 0; i0 < D/2; i0 += WARP_SIZE) {
+                const int i = i0 + threadIdx.x;
+
+                KV_tmp2[k*(D/2) + i].x =  __low2float(V_h2[(k_VKQ_0 + k)*stride_KV2 + i]);
+                KV_tmp2[k*(D/2) + i].y = __high2float(V_h2[(k_VKQ_0 + k)*stride_KV2 + i]);
+            }
+        }
+
+        __syncthreads();
+
+#pragma unroll
+        for (int k = 0; k < FATTN_KQ_STRIDE_TILE_F32; ++k) {
+            float2 V_k[(D/2)/WARP_SIZE];
+            float  KQ_k[ncols/nwarps];
+
+#pragma unroll
+            for (int i0 = 0; i0 < D/2; i0 += WARP_SIZE) {
+                const int i = i0 + threadIdx.x;
+
+                V_k[i0/WARP_SIZE] = KV_tmp2[k*(D/2) + i];
+            }
+#pragma unroll
+            for (int j0 = 0; j0 < ncols; j0 += nwarps) {
+                const int j = j0 + threadIdx.y;
+
+                KQ_k[j0/nwarps] = KQ[j*FATTN_KQ_STRIDE_TILE_F32 + k];
+            }
+
+#pragma unroll
+            for (int i0 = 0; i0 < D/2; i0 += WARP_SIZE) {
+#pragma unroll
+                for (int j0 = 0; j0 < ncols; j0 += nwarps) {
+                    VKQ[j0/nwarps][i0/WARP_SIZE].x += V_k[i0/WARP_SIZE].x*KQ_k[j0/nwarps];
+                    VKQ[j0/nwarps][i0/WARP_SIZE].y += V_k[i0/WARP_SIZE].y*KQ_k[j0/nwarps];
+                }
+            }
+        }
+
+        __syncthreads();
+    }
+
+#pragma unroll
+    for (int j_VKQ_0 = 0; j_VKQ_0 < ncols; j_VKQ_0 += nwarps) {
+        const int j_VKQ = j_VKQ_0 + threadIdx.y;
+
+        float kqsum_j = kqsum[j_VKQ_0/nwarps];
+        kqsum_j = warp_reduce_sum(kqsum_j);
+
+#pragma unroll
+        for (int i00 = 0; i00 < D; i00 += 2*WARP_SIZE) {
+            const int i0 = i00 + 2*threadIdx.x;
+
+            float2 dst_val = VKQ[j_VKQ_0/nwarps][i0/(2*WARP_SIZE)];
+            if (parallel_blocks == 1) {
+                dst_val.x /= kqsum_j;
+                dst_val.y /= kqsum_j;
+            }
+            const int j_dst = (ic0 + j_VKQ)*parallel_blocks + ip;
+            dst[j_dst*D*gridDim.y + D*blockIdx.y + i0 + 0] = dst_val.x;
+            dst[j_dst*D*gridDim.y + D*blockIdx.y + i0 + 1] = dst_val.y;
+        }
+
+        if (parallel_blocks != 1 && threadIdx.x == 0) {
+            dst_meta[(ic0 + j_VKQ)*gridDim.y*parallel_blocks + blockIdx.y*parallel_blocks + ip] = make_float2(kqmax[j_VKQ_0/nwarps], kqsum_j);
+        }
+    }
+}
+
+template <int D, int cols_per_block, int parallel_blocks> void launch_fattn_tile_f32(
+        const ggml_tensor * Q, const ggml_tensor * K, const ggml_tensor * V, ggml_tensor * KQV, const ggml_tensor * mask,
+        ggml_cuda_pool & pool, cudaStream_t main_stream
+) {
+    ggml_cuda_pool_alloc<float>  dst_tmp(pool);
+    ggml_cuda_pool_alloc<float2> dst_tmp_meta(pool);
+
+    if (parallel_blocks > 1) {
+        dst_tmp.alloc(parallel_blocks*ggml_nelements(KQV));
+        dst_tmp_meta.alloc(parallel_blocks*ggml_nrows(KQV));
+    }
+
+    constexpr int  nwarps = 8;
+    const     dim3 block_dim(WARP_SIZE, nwarps, 1);
+    const     dim3 blocks_num(parallel_blocks*((Q->ne[1] + cols_per_block - 1) / cols_per_block), Q->ne[2], Q->ne[3]);
+    const     int  shmem = 0;
+
+    float scale    = 1.0f;
+    float max_bias = 0.0f;
+
+    memcpy(&scale,    (float *) KQV->op_params + 0, sizeof(float));
+    memcpy(&max_bias, (float *) KQV->op_params + 1, sizeof(float));
+
+    const uint32_t n_head      = Q->ne[2];
+    const uint32_t n_head_log2 = 1u << (uint32_t) floorf(log2f((float) n_head));
+
+    const float m0 = powf(2.0f, -(max_bias       ) / n_head_log2);
+    const float m1 = powf(2.0f, -(max_bias / 2.0f) / n_head_log2);
+
+    flash_attn_tile_ext_f32<D, cols_per_block, nwarps, parallel_blocks>
+        <<<blocks_num, block_dim, shmem, main_stream>>> (
+                (const char *) Q->data,
+                (const char *) K->data,
+                (const char *) V->data,
+                mask ? ((const char *) mask->data) : nullptr,
+                parallel_blocks == 1 ? (float *) KQV->data : dst_tmp.ptr, dst_tmp_meta.ptr,
+                scale, max_bias, m0, m1, n_head_log2,
+                Q->ne[0], Q->ne[1], Q->ne[2], Q->ne[3],
+                K->ne[0], K->ne[1], K->ne[2], K->ne[3],
+                mask ? mask->ne[1] : 0, mask ?  mask->nb[1] : 0,
+                Q->nb[1], Q->nb[2], Q->nb[3],
+                K->nb[1], K->nb[2], K->nb[3],
+                KQV->ne[0], KQV->ne[1], KQV->ne[2], KQV->ne[3]
+                );
+    CUDA_CHECK(cudaGetLastError());
+
+    if (parallel_blocks == 1) {
+        return;
+    }
+
+    const dim3 block_dim_combine(D, 1, 1);
+    const dim3 blocks_num_combine(Q->ne[1], blocks_num.y, blocks_num.z);
+    const int  shmem_combine = 0;
+
+    flash_attn_combine_results<D, parallel_blocks>
+        <<<blocks_num_combine, block_dim_combine, shmem_combine, main_stream>>>
+        (dst_tmp.ptr, dst_tmp_meta.ptr, (float *) KQV->data);
+    CUDA_CHECK(cudaGetLastError());
+}
+
+void ggml_cuda_flash_attn_ext_tile_f32(ggml_backend_cuda_context & ctx, ggml_tensor * dst) {
+    const ggml_tensor * Q = dst->src[0];
+    const ggml_tensor * K = dst->src[1];
+    const ggml_tensor * V = dst->src[2];
+
+    const ggml_tensor * mask = dst->src[3];
+
+    ggml_tensor * KQV = dst;
+
+    const int32_t precision = KQV->op_params[2];
+    GGML_ASSERT(precision == GGML_PREC_DEFAULT);
+    GGML_ASSERT(Q->ne[0] == 64 || Q->ne[0] == 128 && "FlashAttention without tensor cores only supports head sizes 64 and 128.");
+
+    if (Q->ne[1] <= 16) {
+        constexpr int cols_per_block = 16;
+        constexpr int parallel_blocks = 4;
+        switch (Q->ne[0]) {
+            case 64:
+                launch_fattn_tile_f32< 64, cols_per_block, parallel_blocks>(Q, K, V, KQV, mask, ctx.pool(), ctx.stream());
+                break;
+            case 128:
+                launch_fattn_tile_f32<128, cols_per_block, parallel_blocks>(Q, K, V, KQV, mask, ctx.pool(), ctx.stream());
+                break;
+            default:
+                GGML_ASSERT(false);
+                break;
+        }
+        return;
+    }
+
+    if (Q->ne[1] <= 32) {
+        constexpr int cols_per_block = 32;
+        constexpr int parallel_blocks = 4;
+        switch (Q->ne[0]) {
+            case 64:
+                launch_fattn_tile_f32< 64, cols_per_block, parallel_blocks>(Q, K, V, KQV, mask, ctx.pool(), ctx.stream());
+                break;
+            case 128:
+                launch_fattn_tile_f32<128, cols_per_block, parallel_blocks>(Q, K, V, KQV, mask, ctx.pool(), ctx.stream());
+                break;
+            default:
+                GGML_ASSERT(false);
+                break;
+        }
+        return;
+    }
+
+    constexpr int cols_per_block = 32;
+    constexpr int parallel_blocks = 1;
+    switch (Q->ne[0]) {
+        case 64:
+            launch_fattn_tile_f32< 64, cols_per_block, parallel_blocks>(Q, K, V, KQV, mask, ctx.pool(), ctx.stream());
+            break;
+        case 128:
+            launch_fattn_tile_f32<128, cols_per_block, parallel_blocks>(Q, K, V, KQV, mask, ctx.pool(), ctx.stream());
+            break;
+        default:
+            GGML_ASSERT(false);
+            break;
+    }
+}

ggml-cuda/fattn-tile-f32.cuh

@@ -0,0 +1,3 @@
+#include "common.cuh"
+
+void ggml_cuda_flash_attn_ext_tile_f32(ggml_backend_cuda_context & ctx, ggml_tensor * dst);

ggml-cuda/fattn.cu

@@ -1,5 +1,7 @@
 #include "common.cuh"
 #include "fattn-common.cuh"
+#include "fattn-tile-f16.cuh"
+#include "fattn-tile-f32.cuh"
 #include "fattn-vec-f16.cuh"
 #include "fattn-vec-f32.cuh"
 #include "fattn.cuh"
@@ -88,7 +90,7 @@ static __global__ void flash_attn_ext_f16(
 
     // ALiBi
     if (max_bias > 0.0f) {
-        const int h = blockIdx.y;
+        const uint32_t h = blockIdx.y;
 
         const float base = h < n_head_log2 ? m0 : m1;
         const int   exph = h < n_head_log2 ? h + 1 : 2*(h - n_head_log2) + 1;
@@ -541,13 +543,31 @@ void ggml_cuda_flash_attn_ext(ggml_backend_cuda_context & ctx, ggml_tensor * dst
 
     const int32_t precision = KQV->op_params[2];
 
+    // On AMD the tile kernels perform poorly, use the vec kernel instead:
+    if (cc >= CC_OFFSET_AMD) {
+        if (precision == GGML_PREC_DEFAULT) {
+            ggml_cuda_flash_attn_ext_vec_f16_no_mma(ctx, dst);
+        } else {
+            ggml_cuda_flash_attn_ext_vec_f32(ctx, dst);
+        }
+        return;
+    }
+
     if (!fast_fp16_available(cc)) {
-        ggml_cuda_flash_attn_ext_vec_f32(ctx, dst);
+        if (Q->ne[1] <= 8) {
+            ggml_cuda_flash_attn_ext_vec_f32(ctx, dst);
+        } else {
+            ggml_cuda_flash_attn_ext_tile_f32(ctx, dst);
+        }
         return;
     }
 
     if (!fp16_mma_available(cc)) {
-        ggml_cuda_flash_attn_ext_vec_f16_no_mma(ctx, dst);
+        if (Q->ne[1] <= 8) {
+            ggml_cuda_flash_attn_ext_vec_f16_no_mma(ctx, dst);
+        } else {
+            ggml_cuda_flash_attn_ext_tile_f16(ctx, dst);
+        }
         return;
     }