[HARDWARE][POWER] optimize math functions of VSX power (#47321)

Signed-off-by: Akash Kaothalkar <akashkaothalkar@akashs-mbp.bl1-in.ibm.com>
Signed-off-by: Akash Kaothalkar <akash.kaothalkar@ibm.com>
Signed-off-by: Akash kaothalkar <akash.kaothalkar@ibm.com>
Signed-off-by: Rukhaiya <bibirukhaiya123@gmail.com>
Co-authored-by: Akash Kaothalkar <akashkaothalkar@akashs-mbp.bl1-in.ibm.com>
Co-authored-by: Akash Kaothalkar <akash.kaothalkar@ibm.com>
This commit is contained in:
Rukhaiya2004
2026-07-07 09:35:47 +00:00
committed by GitHub
co-authored by Akash Kaothalkar Akash Kaothalkar
parent c5b66233b2
commit c85d72076a
5 changed files with 154 additions and 54 deletions
+30 -8
View File
@@ -129,10 +129,12 @@ class SiluAndMul(CustomOp):
def __init__(self, *, compile_native: bool = True):
super().__init__(compile_native=compile_native)
if current_platform.is_cuda_alike() or current_platform.is_xpu():
if (
current_platform.is_cuda_alike()
or current_platform.is_cpu()
or current_platform.is_xpu()
):
self.op = torch.ops._C.silu_and_mul
elif current_platform.is_cpu():
self._forward_method = self.forward_native
@staticmethod
def forward_native(x: torch.Tensor) -> torch.Tensor:
@@ -150,6 +152,11 @@ class SiluAndMul(CustomOp):
def forward_xpu(self, x: torch.Tensor) -> torch.Tensor:
return self.forward_cuda(x)
def forward_cpu(self, x: torch.Tensor) -> torch.Tensor:
if current_platform.get_cpu_architecture() == CpuArchEnum.POWERPC:
return self.forward_cuda(x)
return self.forward_native(x)
@CustomOp.register("silu_and_mul_with_clamp")
class SiluAndMulWithClamp(CustomOp):
@@ -417,14 +424,14 @@ class GeluAndMul(CustomOp):
self.op(out, x)
return out
def forward_cpu(self, x: torch.Tensor) -> torch.Tensor:
if self.op:
return self.forward_cuda(x)
return self.native(x)
def forward_xpu(self, x: torch.Tensor) -> torch.Tensor:
return self.forward_cuda(x)
def forward_cpu(self, x: torch.Tensor) -> torch.Tensor:
if current_platform.get_cpu_architecture() == CpuArchEnum.POWERPC:
return self.forward_cuda(x)
return self.forward_native(x)
def extra_repr(self) -> str:
return f"approximate={repr(self.approximate)}"
@@ -526,6 +533,11 @@ class NewGELU(CustomOp):
def forward_xpu(self, x: torch.Tensor) -> torch.Tensor:
return self.forward_cuda(x)
def forward_cpu(self, x: torch.Tensor) -> torch.Tensor:
if current_platform.get_cpu_architecture() == CpuArchEnum.POWERPC:
return self.forward_cuda(x)
return self.forward_native(x)
# --8<-- [start:gelu_fast]
@CustomOp.register("gelu_fast")
@@ -553,6 +565,11 @@ class FastGELU(CustomOp):
def forward_xpu(self, x: torch.Tensor) -> torch.Tensor:
return self.forward_cuda(x)
def forward_cpu(self, x: torch.Tensor) -> torch.Tensor:
if current_platform.get_cpu_architecture() == CpuArchEnum.POWERPC:
return self.forward_cuda(x)
return self.forward_native(x)
# --8<-- [start:quick_gelu]
@CustomOp.register("quick_gelu")
@@ -581,6 +598,11 @@ class QuickGELU(CustomOp):
def forward_xpu(self, x: torch.Tensor) -> torch.Tensor:
return self.forward_cuda(x)
def forward_cpu(self, x: torch.Tensor) -> torch.Tensor:
if current_platform.get_cpu_architecture() == CpuArchEnum.POWERPC:
return self.forward_cuda(x)
return self.forward_native(x)
# --8<-- [start:relu2]
@CustomOp.register("relu2")