compiler/plugins/target/LLVMCPU/builtins/ukernel/BUILD.bazel

# Copyright 2026 The IREE Authors
#
# Licensed under the Apache License v2.0 with LLVM Exceptions.
# See https://llvm.org/LICENSE.txt for license information.
# SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception

load("//build_tools/bazel:build_defs.oss.bzl", "iree_cmake_extra_content")
load("//build_tools/bazel:iree_bitcode_library.bzl", "iree_bitcode_library")
load("//build_tools/embed_data:build_defs.bzl", "iree_c_embed_data")

package(
    default_visibility = ["//visibility:public"],
    features = ["layering_check"],
    licenses = ["notice"],  # Apache 2.0
)

iree_cmake_extra_content(
    content = """
if(NOT IREE_TARGET_BACKEND_LLVM_CPU)
  return()
endif()
""",
    inline = True,
)

#===------------------------------------------------------------------------===#
# UKernel bitcode files
#===------------------------------------------------------------------------===#
#
# Each entry below compiles one .c file to one .bc file for a specific
# (arch, features) tuple. Each .c file is named after the MMA intrinsic it
# implements (lowercased, with `iree_uk_` prefix), matching the
# verbatim-intrinsic convention used by the AMDGPU C ukernels. The bitcode
# output name appends the build-side `<arch>_<features>` suffix so the
# compiler-side lookup in `EmbeddedDataDirectory` can find the right bitcode
# for a given target's feature set. See README.md for the design.

X86_64_AVX512_BF16_COPTS = [
    "-mavx512f",
    "-mavx512bf16",
]

X86_64_AVX512_VNNI_COPTS = [
    "-mavx512f",
    "-mavx512bw",
    "-mavx512vnni",
]

# (stem, features, copts) for each x86_64 ukernel. `stem` is both the .c file
# base name and the ukernel function name; `features` is the ISA-extension
# suffix appended (as `_x86_64_<features>` to the library name and
# `.x86_64_<features>.bc` to the output) for the EmbeddedDataDirectory lookup.
X86_64_UKERNELS = [
    (
        "iree_uk_mma_x86_avx512bf16_1x16x2_f32_bf16",
        "avx512bf16",
        X86_64_AVX512_BF16_COPTS,
    ),
    (
        "iree_uk_mma_x86_avx512vnni_16x16x2_i32_i8_casti16",
        "avx512vnni",
        X86_64_AVX512_VNNI_COPTS,
    ),
]

[iree_bitcode_library(
    name = "%s_x86_64_%s" % (stem, features),
    srcs = ["%s.c" % stem],
    out = "%s.x86_64_%s.bc" % (stem, features),
    arch = "x86_64",
    copts = copts,
    internal_hdrs = ["common.h"],
) for (stem, features, copts) in X86_64_UKERNELS]

#===------------------------------------------------------------------------===#
# Embedded bitcode TOC
#===------------------------------------------------------------------------===#

iree_c_embed_data(
    name = "iree_uk_cpu_bitcode",
    srcs = [
        ":%s.x86_64_%s.bc" % (stem, features)
        for (stem, features, copts) in X86_64_UKERNELS
    ],
    c_file_output = "iree_uk_cpu_bitcode.c",
    flatten = True,
    h_file_output = "iree_uk_cpu_bitcode.h",
)