Lumpiasty/llama.cpp
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

vulkan/cpu: Support f16 as SET_ROWS src. (#25432)

Browse Source 
* vulkan/cpu: Support f16 as SET_ROWS src.

This adds full support for f16 SET_ROWS (equivalent to f32) to vulkan and CPU
backends, and adds more backend tests.

* Set DenormPreserve 16 when supported, to try to fix failures on Intel

* tune error threshold

* update metal supports_op
This commit is contained in:
Jeff Bolz
2026-07-14 08:26:55 -05:00
committed by GitHub
parent 8ff8c4299d
commit 7cbd61002d
7 changed files with 150 additions and 100 deletions
Download Patch File Download Diff File Expand all files Collapse all files
ggml/src/ggml-metal/ggml-metal-device.m
+5 -1
View File
@@ -1340,7 +1340,11 @@ bool ggml_metal_device_supports_op(ggml_metal_device_t dev, const struct ggml_te
return op->src[0]->type != GGML_TYPE_NVFP4;
case GGML_OP_SET_ROWS:
{
if (op->src[0]->type != GGML_TYPE_F32 && op->src[0]->type != GGML_TYPE_F16) {
if (op->src[0]->type == GGML_TYPE_F16) {
return op->type == GGML_TYPE_F16;
}
if (op->src[0]->type != GGML_TYPE_F32) {
return false;
}