Skip to content

Comments

backport xe3p#4724

Open
dyoussif wants to merge 1 commit intomainfrom
dyoussif/merge_main
Open

backport xe3p#4724
dyoussif wants to merge 1 commit intomainfrom
dyoussif/merge_main

Conversation

@dyoussif
Copy link
Contributor

No description provided.

@dyoussif dyoussif requested review from a team as code owners February 24, 2026 20:51
@github-actions github-actions bot added platform:gpu-intel Codeowner: @oneapi-src/onednn-gpu-intel component:api Codeowner: @oneapi-src/onednn-arch component:graph-api Codeowner: @oneapi-src/onednn-graph component:tests Codeowner: @oneapi-src/onednn-arch component:build component:common third_party labels Feb 24, 2026
@dyoussif dyoussif force-pushed the dyoussif/merge_main branch from 16357f1 to b6e2770 Compare February 24, 2026 20:57
case gpu::intel::compute::gpu_arch_t::xe2:
case gpu_arch_t::xe3p_35_10:
case gpu_arch_t::xe3p_35_11:
case gpu_arch_t::xe3p_35_unknown:
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Nit: same style for prefixes here and one below?

bool mayiuse_systolic_ = false;
bool mayiuse_ngen_kernels_ = false;
bool mayiuse_system_memory_allocators_ = false;
bool is_efficient_64bit_ = false;
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Would be nice to have a comment explaining what is efficient 64 bit...

mov(mod, temp, src1);
mul(mod, dst, src0, temp);
state.ra.safeRelease(tempRange);
}else
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Is this file in clang-format exception list?


state.ra.safeRelease(inputs.ldaq);
state.ra.safeRelease(inputs.ldbq);
//state.ra.safeRelease(inputs.ldcq);
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Debug product?

if (!isBlock2D(astrategy.accessType) && !asIfBlock2D) return 0;
if (hw == HW::Xe2) return 16;
if (hw == HW::Xe3) return 16;
if (hw >= HW::XE3P_35_10) return 4;
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Minor: can it be unsafe for future platforms and lead to potential issues?

return status::success;
}

void debugdump_processed_source(const std::string &source,
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

This function was moved to compute folder, please drop.


if(DNNL_WITH_XE3P)
add_definitions_with_host_compiler(-DXE3P=1)
endif()
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

It feels like this is not needed.

VCHECK_SDP_PRIMITIVE(false, status::unimplemented,
"f32 fused sdpa supported for: causal mask or cases with "
"head_size <= 64, seq_len >= 16");
}
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

This reverts the change that is on main. Please drop changes from this file.

if(DNNL_WITH_XE3P)
add_definitions_with_host_compiler(-DDNNL_WITH_XE3P)
add_definitions_with_host_compiler(-DXE3P)
endif()
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Follow TODO?

EXPECT_EQ(num_inplace_pairs, 0U);
#else
EXPECT_EQ(num_inplace_pairs, engine == dnnl_gpu ? 1U : 0U);
#endif
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Please drop changes from this file, too.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

component:api Codeowner: @oneapi-src/onednn-arch component:build component:common component:graph-api Codeowner: @oneapi-src/onednn-graph component:tests Codeowner: @oneapi-src/onednn-arch platform:gpu-intel Codeowner: @oneapi-src/onednn-gpu-intel third_party

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants