已合并
always use aclnnGeLUV2 in 950 #5316
XianglongZeng创建于 6月30日
always use aclnnGeLUV2 in 950 #5316
已合并
XianglongZeng创建于 6月30日
2 个文件变更+3-3
@@ -28,7 +28,7 @@ at::Tensor gelu_backward(const at::Tensor & grad_output, const at::Tensor & self
28 grad_output.options().dtype(output_dtype_0));28 grad_output.options().dtype(output_dtype_0));
29 auto approximate_str = op_infer::npu_gelu_approximate_str(approximate);29 auto approximate_str = op_infer::npu_gelu_approximate_str(approximate);
30 auto approximate_ptr = const_cast<char *>(approximate_str.c_str());30 auto approximate_ptr = const_cast<char *>(approximate_str.c_str());
31- if (!gelu_sc) {31+ if (!c10_npu::IsAclnnOnly() && !gelu_sc) {
32 DO_COMPATIBILITY(aclnnGeluBackward, acl_op::gelu_backward(grad_output, self, approximate));32 DO_COMPATIBILITY(aclnnGeluBackward, acl_op::gelu_backward(grad_output, self, approximate));
33 EXEC_NPU_CMD(aclnnGeluBackward, grad_output, self, grad_input);33 EXEC_NPU_CMD(aclnnGeluBackward, grad_output, self, grad_input);
34 } else {34 } else {
@@ -21,7 +21,7 @@ using npu_preparation = at_npu::native::OpPreparation;
21at::Tensor& gelu_out(const at::Tensor & self, c10::string_view approximate, at::Tensor& out)21at::Tensor& gelu_out(const at::Tensor & self, c10::string_view approximate, at::Tensor& out)
22{22{
23 static auto gelu_sc = at_npu::native::env::CheckCompatibleImpl();23 static auto gelu_sc = at_npu::native::env::CheckCompatibleImpl();
24- if (!gelu_sc) {24+ if (!c10_npu::IsAclnnOnly() && !gelu_sc) {
25 DO_COMPATIBILITY(aclnnGelu, acl_op::gelu_out(self, approximate, out));25 DO_COMPATIBILITY(aclnnGelu, acl_op::gelu_out(self, approximate, out));
26 EXEC_NPU_CMD(aclnnGelu, self, out);26 EXEC_NPU_CMD(aclnnGelu, self, out);
27 } else {27 } else {
@@ -30,4 +30,4 @@ at::Tensor& gelu_out(const at::Tensor & self, c10::string_view approximate, at::
30 }30 }
31 return out;31 return out;
32}32}
33-}33+}