[Performance] Add Triton kernel for Gemma3n sparse GELU (#48498)

Signed-off-by: BWAAEEEK <jooho414@gmail.com>
Co-authored-by: Andreas Karatzas <akaratza@amd.com>
Co-authored-by: Cyrus Leung <tlleungac@connect.ust.hk>
This commit is contained in:
JooHo Lee
2026-09-14 07:42:15 +00:00
committed by GitHub
co-authored by Andreas Karatzas Cyrus Leung
parent e6b4e47d2d
commit 78e84261ab
15 changed files with 555 additions and 21 deletions
+4 -1
View File
@@ -755,7 +755,10 @@ class CudaPlatformBase(Platform):
rms_norm = ["oink"] + default
return IrOpPriorityConfig.with_default(
default, rms_norm=rms_norm, fused_add_rms_norm=rms_norm
default,
rms_norm=rms_norm,
fused_add_rms_norm=rms_norm,
gelu_and_mul_sparse=["triton", "native"],
)
@classmethod
+4 -1
View File
@@ -1161,7 +1161,10 @@ class RocmPlatform(Platform):
rms_norm = default
return IrOpPriorityConfig.with_default(
default, rms_norm=rms_norm, fused_add_rms_norm=rms_norm
default,
rms_norm=rms_norm,
fused_add_rms_norm=rms_norm,
gelu_and_mul_sparse=["native"],
)
@classmethod
+1 -1
View File
@@ -491,7 +491,7 @@ class XPUPlatform(Platform):
using_inductor = cc.backend == "inductor" and cc.mode != CompilationMode.NONE
default = ["native"] if using_inductor else ["vllm_c", "native"]
return IrOpPriorityConfig.with_default(default)
return IrOpPriorityConfig.with_default(default, gelu_and_mul_sparse=["native"])
@classmethod
def device_count(cls) -> int: