[CPU] Implement MXFP4 Gemm kernels for intel AMX to support GPT OSS series. (#14385)
This commit is contained in:
@@ -45,7 +45,7 @@ namespace {
|
||||
} \
|
||||
}()
|
||||
|
||||
// dispatch: bfloat16, float16, int8_t, fp8_e4m3
|
||||
// dispatch: bfloat16, float16, int8_t, fp8_e4m3, uint8_t(mxfp4/int4)
|
||||
#define CPU_DISPATCH_PACKED_TYPES(TYPE, ...) \
|
||||
[&] { \
|
||||
switch (TYPE) { \
|
||||
@@ -65,6 +65,10 @@ namespace {
|
||||
using packed_t = at::Float8_e4m3fn; \
|
||||
return __VA_ARGS__(); \
|
||||
} \
|
||||
case at::ScalarType::Byte: { \
|
||||
using packed_t = uint8_t; \
|
||||
return __VA_ARGS__(); \
|
||||
} \
|
||||
default: \
|
||||
TORCH_CHECK(false, "Unsupported floating data type.\n"); \
|
||||
} \
|
||||
|
||||
Reference in New Issue
Block a user