From b3bc90463807a97763c2f1e4082bd9925f4dbc7f Mon Sep 17 00:00:00 2001 From: Yash Raj Pandey <55940078+devYRPauli@users.noreply.github.com> Date: Mon, 10 Aug 2026 08:01:44 -0400 Subject: [PATCH] ggml : require contiguous src for ROLL on CUDA and Metal (llama/25928) ggml_roll only asserts nb[0] == ggml_type_size, so a permuted src is a valid input, but the CUDA and Metal roll kernels index by ne alone and never read the nb strides. A non-contiguous src therefore produced silently wrong results. Neither backend declared a contiguity requirement in supports_op, so the scheduler did not fall back to the CPU implementation, which does handle strides correctly. Add the requirement to both backends, matching the existing GGML_OP_ROPE guard, and add a permuted test_roll case. --- ggml/src/ggml-cuda/ggml-cuda.cu | 2 +- ggml/src/ggml-metal/ggml-metal-device.m | 3 ++- 2 files changed, 3 insertions(+), 2 deletions(-) diff --git a/ggml/src/ggml-cuda/ggml-cuda.cu b/ggml/src/ggml-cuda/ggml-cuda.cu index dec619324..05e8d7f73 100644 --- a/ggml/src/ggml-cuda/ggml-cuda.cu +++ b/ggml/src/ggml-cuda/ggml-cuda.cu @@ -5185,7 +5185,7 @@ static bool ggml_backend_cuda_device_supports_op(ggml_backend_dev_t dev, const g return max_bias == 0.0f; } case GGML_OP_ROLL: - if(op->src[0]->type == GGML_TYPE_F32) { + if(op->src[0]->type == GGML_TYPE_F32 && ggml_is_contiguous(op->src[0])) { return true; } return false; diff --git a/ggml/src/ggml-metal/ggml-metal-device.m b/ggml/src/ggml-metal/ggml-metal-device.m index 2dc6eb8fd..df85ab02c 100644 --- a/ggml/src/ggml-metal/ggml-metal-device.m +++ b/ggml/src/ggml-metal/ggml-metal-device.m @@ -1268,8 +1268,9 @@ bool ggml_metal_device_supports_op(ggml_metal_device_t dev, const struct ggml_te case GGML_OP_ARGSORT: case GGML_OP_TOP_K: case GGML_OP_ARANGE: - case GGML_OP_ROLL: return true; + case GGML_OP_ROLL: + return ggml_is_contiguous(op->src[0]); case GGML_OP_FLASH_ATTN_EXT: // for new head sizes, add checks here if (op->src[0]->ne[0] != 32 &&