metal : support argsort for ne00 > 1024 (#17247)

* metal : refactor argsort

* cont : sort chunks

* cont : merge sorted buckets

* cont : cleanup
This commit is contained in:
Georgi Gerganov
2025-11-14 09:36:06 +02:00
committed by GitHub
parent 2606b0adab
commit 45c6ef7307
8 changed files with 266 additions and 45 deletions
-2
View File
@@ -904,8 +904,6 @@ bool ggml_metal_device_supports_op(ggml_metal_device_t dev, const struct ggml_te
case GGML_OP_LEAKY_RELU:
return op->src[0]->type == GGML_TYPE_F32;
case GGML_OP_ARGSORT:
// TODO: Support arbitrary column width
return op->src[0]->ne[0] <= 1024;
case GGML_OP_ARANGE:
return true;
case GGML_OP_FLASH_ATTN_EXT: