RoPE: fix back, CUDA support for back + noncont. (ggml-org#11240)

* RoPE: fix back, CUDA support for back + noncont.

* fix comments reg. non-cont. RoPE support [no-ci]

Author: JohannesGaessler

ggml/include/ggml.h

@@ -1500,7 +1500,7 @@ extern "C" {
 
     // rotary position embedding backward, i.e compute dx from dy
     // a - dy
-    GGML_API struct ggml_tensor * ggml_rope_back(
+    GGML_API struct ggml_tensor * ggml_rope_ext_back(
             struct ggml_context * ctx,
             struct ggml_tensor  * a, // gradients of ggml_rope result
             struct ggml_tensor  * b, // positions
@@ -1515,6 +1515,23 @@ extern "C" {
             float                 beta_fast,
             float                 beta_slow);
 
+    GGML_API struct ggml_tensor * ggml_rope_multi_back(
+            struct ggml_context * ctx,
+            struct ggml_tensor  * a,
+            struct ggml_tensor  * b,
+            struct ggml_tensor  * c,
+            int                   n_dims,
+            int                   sections[4],
+            int                   mode,
+            int                   n_ctx_orig,
+            float                 freq_base,
+            float                 freq_scale,
+            float                 ext_factor,
+            float                 attn_factor,
+            float                 beta_fast,
+            float                 beta_slow);
+
+
     // clamp
     // in-place, returns view(a)
     GGML_API struct ggml_tensor * ggml_clamp(

ggml/src/ggml-cpu/ggml-cpu.c

@@ -13668,6 +13668,7 @@ struct ggml_cplan ggml_graph_plan(
                     } break;
                 case GGML_OP_SOFT_MAX:
                 case GGML_OP_ROPE:
+                case GGML_OP_ROPE_BACK:
                     {
                         cur = ggml_type_size(GGML_TYPE_F32) * node->ne[0] * n_tasks;
                     } break;

ggml/src/ggml-cpu/ggml-cpu.cpp

@@ -403,8 +403,6 @@ static bool ggml_backend_cpu_device_supports_op(ggml_backend_dev_t dev, const st
                 op->type != GGML_TYPE_IQ1_M; // missing type_traits.from_float
         case GGML_OP_MUL_MAT:
             return src1->type == GGML_TYPE_F32 || src1->type == ggml_get_type_traits_cpu(src0->type)->vec_dot_type;
-        case GGML_OP_ROPE_BACK:
-            return op->src[2] == NULL && (op->op_params[2] & 4) == 0;
         case GGML_OP_IM2COL_BACK:
             return src0->type == GGML_TYPE_F32 && src1->type == GGML_TYPE_F32;
         case GGML_OP_OUT_PROD:

ggml/src/ggml-cuda/ggml-cuda.cu

@@ -2141,6 +2141,9 @@ static bool ggml_cuda_compute_forward(ggml_backend_cuda_context & ctx, struct gg
         case GGML_OP_ROPE:
             ggml_cuda_op_rope(ctx, dst);
             break;
+        case GGML_OP_ROPE_BACK:
+            ggml_cuda_op_rope_back(ctx, dst);
+            break;
         case GGML_OP_IM2COL:
             ggml_cuda_op_im2col(ctx, dst);
             break;
@@ -3025,7 +3028,11 @@ static bool ggml_backend_cuda_device_supports_op(ggml_backend_dev_t dev, const g
         case GGML_OP_SOFT_MAX:
             return true;
         case GGML_OP_ROPE:
-            return ggml_is_contiguous(op->src[0]);
+        case GGML_OP_ROPE_BACK: {
+            const size_t ts = ggml_type_size(op->src[0]->type);
+            const int64_t ne0_012 = op->src[0]->ne[0] * op->src[0]->ne[1] * op->src[0]->ne[2];
+            return op->src[0]->nb[0] == ts && op->src[0]->nb[3] == ne0_012*ts;
+        }
         case GGML_OP_IM2COL:
         case GGML_OP_POOL_2D:
         case GGML_OP_SUM:
@@ -3081,6 +3088,7 @@ static int64_t get_op_batch_size(const ggml_tensor * op) {
             return op->ne[1];
         case GGML_OP_MUL_MAT_ID:
         case GGML_OP_ROPE:
+        case GGML_OP_ROPE_BACK:
             return op->ne[2];
         default:
             return ggml_nrows(op);