ggml : group all experts in a single ggml_mul_mat_id

cuda : improve mmid row copy

Author: slaren

ggml-cuda.cu

@@ -1231,7 +1231,7 @@ static void ggml_cuda_op_mul_mat_cublas(
 
     if (compute_capability >= CC_VOLTA && (src0->type == GGML_TYPE_F16 || ggml_is_quantized(src0->type)) && ggml_is_contiguous(src0) && row_diff == src0->ne[1] && dst->op_params[0] == GGML_PREC_DEFAULT) {
         // convert src0 and src1 to fp16, multiply as fp16, convert dst to fp32
-        ggml_cuda_pool_alloc<half> src0_as_f16(ctx.pool());
+        ggml_cuda_pool_alloc<half> src0_as_f16(ctx.pool(id));
         if (src0->type != GGML_TYPE_F16) {
             const to_fp16_cuda_t to_fp16_cuda = ggml_get_to_fp16_cuda(src0->type);
             GGML_ASSERT(to_fp16_cuda != nullptr);
@@ -1241,7 +1241,7 @@ static void ggml_cuda_op_mul_mat_cublas(
         }
         const half * src0_ptr = src0->type == GGML_TYPE_F16 ? (const half *) src0_dd_i : src0_as_f16.get();
 
-        ggml_cuda_pool_alloc<half> src1_as_f16(ctx.pool());
+        ggml_cuda_pool_alloc<half> src1_as_f16(ctx.pool(id));
         if (src1->type != GGML_TYPE_F16) {
             const to_fp16_cuda_t to_fp16_cuda = ggml_get_to_fp16_cuda(src1->type);
             GGML_ASSERT(to_fp16_cuda != nullptr);
@@ -1250,7 +1250,7 @@ static void ggml_cuda_op_mul_mat_cublas(
             to_fp16_cuda(src1_ddf_i, src1_as_f16.get(), ne, stream);
         }
         const half * src1_ptr = src1->type == GGML_TYPE_F16 ? (const half *) src1_ddf_i : src1_as_f16.get();
-        ggml_cuda_pool_alloc<half> dst_f16(ctx.pool(), row_diff*src1_ncols);
+        ggml_cuda_pool_alloc<half> dst_f16(ctx.pool(id), row_diff*src1_ncols);
 
         const half alpha_f16 = 1.0f;
         const half beta_f16 = 0.0f;
@@ -1960,20 +1960,84 @@ static void ggml_cuda_mul_mat(ggml_backend_cuda_context & ctx, const ggml_tensor
     }
 }
 
+struct mmid_row_mapping {
+    int64_t i1;
+    int64_t i2;
+};
+
+static __global__ void k_copy_src1_to_contiguous(const char * src1_original, char * src1_contiguous,
+                                                 int * cur_src1_row, mmid_row_mapping * row_mapping,
+                                                 const char * ids_dev, int64_t i02, int64_t ids_nb1, int64_t ids_nb0,
+                                                 int64_t ids_ne1, int64_t n_ids,
+                                                 int64_t ne11,
+                                                 size_t nb11, size_t nb12) {
+    int64_t iid1 = blockIdx.x;
+    int64_t id = blockIdx.y;
+
+    if (iid1 >= ids_ne1 || id >= n_ids) {
+        return;
+    }
+
+    const int32_t row_id_i = *(const int32_t *) (ids_dev + iid1*ids_nb1 + id*ids_nb0);
+
+    if (row_id_i != i02) {
+        return;
+    }
+
+    const int64_t i11 = id % ne11;
+    const int64_t i12 = iid1;
+
+    __shared__ int src1_row;
+    if (threadIdx.x == 0) {
+        src1_row = atomicAdd(cur_src1_row, 1);
+        row_mapping[src1_row] = {id, iid1};
+    }
+    __syncthreads();
+
+    const char * src1_row_original = src1_original + i11*nb11 + i12*nb12;
+    char * src1_row_contiguous = src1_contiguous + src1_row*nb11;
+
+    for (int i = threadIdx.x; i < nb11; i += blockDim.x) {
+        src1_row_contiguous[i] = src1_row_original[i];
+    }
+}
+
+static __global__ void k_copy_dst_from_contiguous(char * dst_original, const char * dst_contiguous,
+                                                  const mmid_row_mapping * row_mapping,
+                                                  int64_t n_rows,
+                                                  int64_t nb1, int64_t nb2) {
+    int64_t i = blockIdx.x;
+
+    if (i >= n_rows) {
+        return;
+    }
+
+    const int64_t i1 = row_mapping[i].i1;
+    const int64_t i2 = row_mapping[i].i2;
+
+    const char * dst_row_contiguous = dst_contiguous + i*nb1;
+    char * dst_row_original = dst_original + i1*nb1 + i2*nb2;
+
+    for (int j = threadIdx.x; j < nb1; j += blockDim.x) {
+        dst_row_original[j] = dst_row_contiguous[j];
+    }
+}
+
+//#define MMID_MEMCPY
+
 static void ggml_cuda_mul_mat_id(ggml_backend_cuda_context & ctx, ggml_tensor * dst) {
     const ggml_tensor * src0 = dst->src[0];
     const ggml_tensor * src1 = dst->src[1];
     const ggml_tensor * ids  = dst->src[2];
 
+    GGML_TENSOR_BINARY_OP_LOCALS
+
     GGML_ASSERT(!ggml_backend_buffer_is_cuda_split(src0->buffer) && "mul_mat_id does not support split buffers");
 
     cudaStream_t stream = ctx.stream();
 
-    const size_t nb11 = src1->nb[1];
-    const size_t nb1  =  dst->nb[1];
-
-    const int32_t id = ((int32_t *) dst->op_params)[0];
-    const int32_t n_as = src0->ne[2];
+    const int64_t n_as = ne02;
+    const int64_t n_ids = ids->ne[0];
 
     std::vector<char> ids_host(ggml_nbytes(ids));
     const char * ids_dev = (const char *) ids->data;
@@ -1982,27 +2046,47 @@ static void ggml_cuda_mul_mat_id(ggml_backend_cuda_context & ctx, ggml_tensor *
 
     ggml_tensor src0_row = *src0;
     ggml_tensor src1_row = *src1;
-    ggml_tensor dst_row = *dst;
+    ggml_tensor dst_row  = *dst;
 
     char * src0_original = (char *) src0->data;
     char * src1_original = (char *) src1->data;
     char * dst_original  = (char *)  dst->data;
 
     src0_row.ne[2] = 1;
     src0_row.ne[3] = 1;
-    src0_row.nb[3] = src0->nb[2];
+    src0_row.nb[3] = nb02;
 
-    if (src1->ne[1] == 1) {
-        for (int64_t i01 = 0; i01 < ids->ne[1]; i01++) {
-            const int32_t row_id = *(const int32_t *) (ids_host.data() + i01*ids->nb[1] + id*ids->nb[0]);
+    src1_row.ne[1] = 1;
+    src1_row.ne[2] = 1;
+    src1_row.ne[3] = 1;
+    src1_row.nb[2] = nb11;
+    src1_row.nb[3] = nb11;
 
-            GGML_ASSERT(row_id >= 0 && row_id < n_as);
+    dst_row.ne[1] = 1;
+    dst_row.ne[2] = 1;
+    dst_row.ne[3] = 1;
+    dst_row.nb[2] = nb1;
+    dst_row.nb[3] = nb1;
 
-            src0_row.data = src0_original + row_id*src0->nb[2];
-            src1_row.data = src1_original + i01*src1->nb[1];
-            dst_row.data  =  dst_original + i01*dst->nb[1];
+    if (ne12 == 1) {
+        for (int64_t iid1 = 0; iid1 < ids->ne[1]; iid1++) {
+            for (int64_t id = 0; id < n_ids; id++) {
+                const int32_t i02 = *(const int32_t *) (ids_host.data() + iid1*ids->nb[1] + id*ids->nb[0]);
 
-            ggml_cuda_mul_mat(ctx, &src0_row, &src1_row, &dst_row);
+                GGML_ASSERT(i02 >= 0 && i02 < n_as);
+
+                const int64_t i11 = id % ne11;
+                const int64_t i12 = iid1;
+
+                const int64_t i1 = id;
+                const int64_t i2 = i12;
+
+                src0_row.data = src0_original + i02*nb02;
+                src1_row.data = src1_original + i11*nb11 + i12*nb12;
+                dst_row.data  =  dst_original + i1*nb1   + i2*nb2;
+
+                ggml_cuda_mul_mat(ctx, &src0_row, &src1_row, &dst_row);
+            }
         }
     } else {
         ggml_cuda_pool_alloc<char> src1_contiguous(ctx.pool(), sizeof(float)*ggml_nelements(src1));
@@ -2011,55 +2095,104 @@ static void ggml_cuda_mul_mat_id(ggml_backend_cuda_context & ctx, ggml_tensor *
         src1_row.data = src1_contiguous.get();
         dst_row.data  =  dst_contiguous.get();
 
-        for (int32_t row_id = 0; row_id < n_as; ++row_id) {
+        for (int64_t i02 = 0; i02 < n_as; i02++) {
             int64_t num_src1_rows = 0;
-            for (int64_t i01 = 0; i01 < ids->ne[1]; i01++) {
-                const int32_t row_id_i = *(const int32_t *) (ids_host.data() + i01*ids->nb[1] + id*ids->nb[0]);
 
-                if (row_id_i != row_id) {
-                    continue;
-                }
+            for (int64_t iid1 = 0; iid1 < ids->ne[1]; iid1++) {
+                for (int64_t id = 0; id < n_ids; id++) {
+                    const int32_t row_id_i = *(const int32_t *) (ids_host.data() + iid1*ids->nb[1] + id*ids->nb[0]);
+
+                    if (row_id_i != i02) {
+                        continue;
+                    }
 
-                GGML_ASSERT(row_id >= 0 && row_id < n_as);
+                    GGML_ASSERT(i02 >= 0 && i02 < n_as);
 
-                CUDA_CHECK(cudaMemcpyAsync(src1_contiguous.get() + num_src1_rows*nb11, src1_original + i01*nb11,
-                                        nb11, cudaMemcpyDeviceToDevice, stream));
-                num_src1_rows++;
+#ifdef MMID_MEMCPY
+                    const int64_t i11 = id % ne11;
+                    const int64_t i12 = iid1;
+                    CUDA_CHECK(cudaMemcpyAsync(src1_contiguous.get() + num_src1_rows*nb11,
+                                               src1_original + i11*nb11 + i12*nb12,
+                                               nb11, cudaMemcpyDeviceToDevice, stream));
+#endif
+                    num_src1_rows++;
+                }
             }
 
             if (num_src1_rows == 0) {
                 continue;
             }
 
-            src0_row.data = src0_original + row_id*src0->nb[2];
+#ifndef MMID_MEMCPY
+            ggml_cuda_pool_alloc<int> dev_cur_src1_row(ctx.pool(), 1);
+            ggml_cuda_pool_alloc<mmid_row_mapping> dev_row_mapping(ctx.pool(), num_src1_rows);
+            CUDA_CHECK(cudaMemsetAsync(dev_cur_src1_row.get(), 0, sizeof(int), stream));
 
-            src1_row.ne[1] = num_src1_rows;
-            dst_row.ne[1] = num_src1_rows;
+            {
+                dim3 block_dims(std::min((uint)nb11, 1024u));
+                dim3 grid_dims(ids->ne[1], n_ids);
+                k_copy_src1_to_contiguous<<<grid_dims, block_dims, 0, stream>>>(
+                        src1_original, src1_contiguous.get(),
+                        dev_cur_src1_row.get(), dev_row_mapping.get(),
+                        ids_dev, i02, ids->nb[1], ids->nb[0],
+                        ids->ne[1], n_ids,
+                        ne11,
+                        nb11, nb12);
+                CUDA_CHECK(cudaGetLastError());
+            }
+#endif
+
+            src0_row.data = src0_original + i02*nb02;
 
+            GGML_ASSERT(nb11 == sizeof(float)*ne10);
+            GGML_ASSERT(nb1 == sizeof(float)*ne0);
+
+            src1_row.ne[1] = num_src1_rows;
             src1_row.nb[1] = nb11;
             src1_row.nb[2] = num_src1_rows*nb11;
             src1_row.nb[3] = num_src1_rows*nb11;
 
+            dst_row.ne[1] = num_src1_rows;
             dst_row.nb[1] = nb1;
             dst_row.nb[2] = num_src1_rows*nb1;
             dst_row.nb[3] = num_src1_rows*nb1;
 
             ggml_cuda_mul_mat(ctx, &src0_row, &src1_row, &dst_row);
 
+#ifndef MMID_MEMCPY
+            {
+                dim3 block_dims(std::min((uint)nb1, 1024u));
+                dim3 grid_dims(num_src1_rows);
+                k_copy_dst_from_contiguous<<<grid_dims, block_dims, 0, stream>>>(
+                        dst_original, dst_contiguous.get(),
+                        dev_row_mapping.get(),
+                        num_src1_rows, nb1, nb2);
+                CUDA_CHECK(cudaGetLastError());
+            }
+#endif
+
+#ifdef MMID_MEMCPY
             num_src1_rows = 0;
-            for (int64_t i01 = 0; i01 < ids->ne[1]; i01++) {
-                const int32_t row_id_i = *(const int32_t *) (ids_host.data() + i01*ids->nb[1] + id*ids->nb[0]);
+            for (int64_t iid1 = 0; iid1 < ids->ne[1]; iid1++) {
+                for (int64_t id = 0; id < n_ids; id++) {
+                    const int32_t row_id_i = *(const int32_t *) (ids_host.data() + iid1*ids->nb[1] + id*ids->nb[0]);
 
-                if (row_id_i != row_id) {
-                    continue;
-                }
+                    if (row_id_i != i02) {
+                        continue;
+                    }
 
-                GGML_ASSERT(row_id >= 0 && row_id < n_as);
+                    GGML_ASSERT(i02 >= 0 && i02 < n_as);
 
-                CUDA_CHECK(cudaMemcpyAsync(dst_original + i01*nb1, dst_contiguous.get() + num_src1_rows*nb1,
-                                        nb1, cudaMemcpyDeviceToDevice, stream));
-                num_src1_rows++;
+                    const int64_t i1 = id;
+                    const int64_t i2 = iid1;
+
+                    CUDA_CHECK(cudaMemcpyAsync(dst_original + i1*nb1 + i2*nb2,
+                                               dst_contiguous.get() + num_src1_rows*nb1,
+                                               nb1, cudaMemcpyDeviceToDevice, stream));
+                    num_src1_rows++;
+                }
             }
+#endif
         }
     }
 }
@@ -2487,7 +2620,8 @@ GGML_CALL static bool ggml_backend_cuda_supports_op(ggml_backend_t backend, cons
 GGML_CALL static bool ggml_backend_cuda_offload_op(ggml_backend_t backend, const ggml_tensor * op) {
     const int min_batch_size = 32;
 
-    return op->ne[1] >= min_batch_size && op->op != GGML_OP_GET_ROWS;
+    return (op->ne[1] >= min_batch_size && op->op != GGML_OP_GET_ROWS) ||
+           (op->ne[2] >= min_batch_size && op->op == GGML_OP_MUL_MAT_ID);
 
     GGML_UNUSED(backend);
 }

ggml-cuda/convert.cu

@@ -45,6 +45,8 @@ static __global__ void dequantize_block_q8_0_f16(const void * __restrict__ vx, h
         vals[ix] = x0[ix];
     }
 
+    __syncthreads();
+
 #pragma unroll
     for (int iy = 0; iy < CUDA_Q8_0_NE_ALIGN; iy += 2*WARP_SIZE) {
         if (need_check && i0 + iy + 2*threadIdx.x >= k) {