From: leonardHONG <redacted>
Date: Mon, 15 Jun 2026 06:11:00 +0000 (+0800)
Subject: CUDA: only support F32/F16 for GGML_OP_REPEAT (llama/24533)
X-Git-Tag: upstream/1.9.2~187
X-Git-Url: https://git.djapps.eu/?a=commitdiff_plain;h=7349e5ae11422064511c1b1f0919b9e4fcadb66e;p=pkg%2Fggml%2Fsources%2Fwhisper.cpp

CUDA: only support F32/F16 for GGML_OP_REPEAT (llama/24533)
---

diff --git a/ggml/src/ggml-cuda/ggml-cuda.cu b/ggml/src/ggml-cuda/ggml-cuda.cu
index 61041bdc..01c29a8c 100644
--- a/ggml/src/ggml-cuda/ggml-cuda.cu
+++ b/ggml/src/ggml-cuda/ggml-cuda.cu
@@ -5337,8 +5337,9 @@ static bool ggml_backend_cuda_device_supports_op(ggml_backend_dev_t dev, const g
             } break;
         case GGML_OP_REPEAT:
             {
+                // the CUDA REPEAT path only implements F32/F16; other types assert at runtime
                 ggml_type src0_type = op->src[0]->type;
-                return src0_type != GGML_TYPE_I32 && src0_type != GGML_TYPE_I16;
+                return src0_type == GGML_TYPE_F32 || src0_type == GGML_TYPE_F16;
             } break;
         case GGML_OP_REPEAT_BACK:
                 return op->type == GGML_TYPE_F32 && (op->src[0]->ne[2]*op->src[0]->ne[3]) <= (1 << 15);