refactor: preserve INT8 APIs through backend extensions

This commit is contained in:
leejet committed 2026-11-01 01:55:51 +08:00
1 parent 051b49751e
commit f0204a7eaa
20 files changed
+1579 -108

No files matched your search

-8
View File
@@ -168,10 +168,6 @@ namespace sd_cpu {
d.prepare = i8_prepare;
d.execute = i8_execute;
GGML_ASSERT(ggml_ext_register_kernel(device, &d));
d.name = "sd.linear_i8.cpu.override.v1";
d.op = GGML_OP_MUL_MAT;
d.custom_op = nullptr;
GGML_ASSERT(ggml_ext_register_kernel(device, &d));
d.name = "sd.quantize_i8_convrot.cpu.v1";
d.op = GGML_OP_CUSTOM;
d.custom_op = sd_ggml_convrot_op();
@@ -179,10 +175,6 @@ namespace sd_cpu {
d.prepare = convrot_prepare;
d.execute = convrot_execute;
GGML_ASSERT(ggml_ext_register_kernel(device, &d));
d.name = "sd.quantize_i8_convrot.cpu.override.v1";
d.op = GGML_OP_QUANTIZE_I8_CONVROT;
d.custom_op = nullptr;
GGML_ASSERT(ggml_ext_register_kernel(device, &d));
}
}