sci-ml/caffe2: drop 2.8.0-r2

Signed-off-by: Alfredo Tupone <tupone@gentoo.org>
This commit is contained in:
Alfredo Tupone
2026-02-02 10:39:28 +01:00
parent 7f133bb0a2
commit b09dd88412
3 changed files with 0 additions and 527 deletions
-2
View File
@@ -1,6 +1,4 @@
DIST composable_kernel-7fe50dc3.tar.gz 5380728 BLAKE2B c89c346d8e2d7a93a9cf26409e477fcdd25c43bc3f99d904c3bfe1bc282c6844ef2f2c80aceabe3bf4494db3457285384d5de5a22281aa426ba7479af82b0caf SHA512 a62f92e2dd7da944bd34bab6cf3bf624f630dc316d29c755e9fd523343c3f7648b0b7e0c9a0c8f5e9654477599ae8be9dac687d4054b0390f064ac2e40fc1cd3
DIST composable_kernel-8086bbe3.tar.gz 4418862 BLAKE2B b710e3d4586899443ec01044dad19fd2f992c351e2f65ba526dfcc47cc65c095beaf8ac21a8f71c02a0eb524d364e817b27241a9198884f2bdae9924b51e24e4 SHA512 8410b5a1c864d71f3034ef0d9d1245078856d09cc191faec59856c229bf11d89ae291036d735cb5cec4f1d72e6e9e8f6921833147f9619d30cfab8722d3a9f63
DIST flash-attention-2.7.4.gh.tar.gz 5841323 BLAKE2B 432999d763f2b3d732580ddfea5d3e01370351db0656546259a5e500a07516dd03c98828bfb55855dabe4adc651033b5d97ea4725ca46158b9970f0fbc662710 SHA512 05a4afb09e666f7404d6a3f8b5256e7bed6eba60a6f1bde2b7dbb96d318975f0b458c2521c7a38d88e97b6e4c27f29077cf787849daf82586e33f43a3d9a84b3
DIST pytorch-2.10.0.tar.gz 62555251 BLAKE2B 6b48b4d3d3802a82d37231c472032f8c9390bdbfe9e810ff811f5521737b46794db51e87f7aad3928be917923ae8ac9c619b14d8d0cfacfc115dffb38347bd43 SHA512 929b1be42954f22e2091e0696d6175c8f30752925ce0bbe3a60a9393aff1f0afb228fa5efaf1ce26e78cd9e99294d150109e9ba9cb3243c90e9ec06cc082f74d
DIST pytorch-2.8.0.tar.gz 56565754 BLAKE2B a8f07513b92f9293f8322508f9fc73a462f89fe51cb1f280af371cee19cbe7e2bf900ba2b3c43fd08ea415566db441a6d6310d77f18477e957641be311a361a5 SHA512 448e9dad4aa10f1793d35e6ffe9f0f69b7719d41e6eccceb687a8d0c148e22d03e4f76170a05308ef9323a7aea41aa74605077ae1d68c6d949f13b3340ebf310
DIST pytorch-2.9.1.tar.gz 55764697 BLAKE2B b22e154034f8a25aa3ef949eb6b0456777e11fe5f97de56c6112d93a2e154db425e97848911af458d179f03d7154956f53b715c7b9d7e7f074e0baceac35dad8 SHA512 d7098408d44e0fee9ded4afd6622df6f08757bf02eee878ae25b62a275f82eb16f96a07027c670c6ffdd431c8714c569249bd8518ac8828a504e99908b8c38b1
-384
View File
@@ -1,384 +0,0 @@
# Copyright 2022-2025 Gentoo Authors
# Distributed under the terms of the GNU General Public License v2
EAPI=8
PYTHON_COMPAT=( python3_{11..14} )
ROCM_VERSION=6.1
inherit python-single-r1 cmake cuda flag-o-matic prefix rocm toolchain-funcs
MYPN=pytorch
MYP=${MYPN}-${PV}
# caffe2-2.6.0 depends on future version of composable kernel
# TODO: replace it with RDEPEND in the future
CK_COMMIT=8086bbe3a78d931eb96fe12fdc014082e18d18d3
CK_P=composable_kernel-${CK_COMMIT:0:8}
FLASH_PV=2.7.4
FLASH_PN=flash-attention
FLASH_P=${FLASH_PN}-${FLASH_PV}
FLASH_ATT_URI="https://github.com/Dao-AILab/${FLASH_PN}/archive/refs/tags/v${FLASH_PV}.tar.gz -> ${FLASH_P}.gh.tar.gz"
AOTRITON_PV=0.9.2b
AOTRITON_PN=aotriton
AOTRITON_P=${AOTRITON_PN}-${AOTRITON_PV}
AOTRITON_tar=${AOTRITON_P}-manylinux_2_28_x86_64-rocm6.3-shared.tar.gz
DESCRIPTION="A deep learning framework"
HOMEPAGE="https://pytorch.org/"
SRC_URI="
https://github.com/pytorch/${MYPN}/archive/refs/tags/v${PV}.tar.gz -> ${MYP}.tar.gz
rocm? (
https://github.com/ROCm/composable_kernel/archive/${CK_COMMIT}.tar.gz
-> ${CK_P}.tar.gz
)
flash? ( ${FLASH_ATT_URI} )
memefficient? ( ${FLASH_ATT_URI} )
"
S="${WORKDIR}"/${MYP}
LICENSE="BSD"
SLOT="0"
KEYWORDS="~amd64 ~arm64"
IUSE="cuda cusparselt distributed fbgemm flash gloo memefficient mkl mpi nnpack +numpy
onednn openblas opencl openmp qnnpack rocm xnnpack"
RESTRICT="test"
REQUIRED_USE="
${PYTHON_REQUIRED_USE}
mpi? ( distributed )
gloo? ( distributed )
?? ( cuda rocm )
rocm? (
|| ( ${ROCM_REQUIRED_USE} )
!flash
)
"
RDEPEND="
${PYTHON_DEPS}
dev-cpp/abseil-cpp:=
dev-cpp/gflags:=
>=dev-cpp/glog-0.5.0
dev-cpp/nlohmann_json
dev-cpp/opentelemetry-cpp
dev-libs/cpuinfo
dev-libs/libfmt:=
dev-libs/protobuf:=
dev-libs/pthreadpool
dev-libs/sleef
sci-ml/foxi
~sci-ml/kineto-0.4.0_p20250617
sci-ml/onnx
virtual/lapack
cuda? (
dev-libs/cudnn
>=sci-ml/cudnn-frontend-1.12.0:=
>=dev-util/nvidia-cuda-toolkit-12.9:=[profiler]
cusparselt? ( dev-libs/cusparselt )
)
fbgemm? ( sci-ml/FBGEMM )
gloo? ( <=sci-ml/gloo-2025.06.04[cuda?] )
mpi? ( virtual/mpi )
nnpack? ( sci-ml/NNPACK )
numpy? ( $(python_gen_cond_dep '
dev-python/numpy[${PYTHON_USEDEP}]
') )
onednn? ( =sci-ml/oneDNN-3.5* )
opencl? ( virtual/opencl )
qnnpack? (
!sci-libs/QNNPACK
sci-ml/gemmlowp
)
rocm? (
>=dev-libs/rccl-6.1 <dev-libs/rccl-6.5
>=dev-util/hip-6.1 <dev-util/hip-6.5
>=dev-util/roctracer-6.1 <dev-util/roctracer-6.5
>=sci-libs/hipBLAS-6.1 <sci-libs/hipBLAS-6.5
>=sci-libs/hipBLASLt-6.1 <sci-libs/hipBLASLt-6.5
>=sci-libs/hipCUB-6.1 <sci-libs/hipCUB-6.5
>=sci-libs/hipFFT-6.1 <sci-libs/hipFFT-6.5
>=sci-libs/hipRAND-6.1 <sci-libs/hipRAND-6.5
>=sci-libs/hipSOLVER-6.1 <sci-libs/hipSOLVER-6.5
>=sci-libs/hipSPARSE-6.1 <sci-libs/hipSPARSE-6.5
>=sci-libs/miopen-6.1 <sci-libs/miopen-6.5
>=sci-libs/rocPRIM-6.1 <sci-libs/rocPRIM-6.5
>=sci-libs/rocThrust-6.1 <sci-libs/rocThrust-6.5
memefficient? ( sci-libs/aotriton-bin:0/0.9 )
)
distributed? (
sci-ml/tensorpipe[cuda?]
dev-cpp/cpp-httplib
)
xnnpack? ( >=sci-ml/XNNPACK-2024.11 )
mkl? ( sci-libs/mkl )
openblas? ( sci-libs/openblas )
"
DEPEND="
${RDEPEND}
dev-libs/flatbuffers
dev-libs/FXdiv
dev-libs/pocketfft
dev-libs/psimd
sci-ml/FP16
$(python_gen_cond_dep '
dev-python/pybind11[${PYTHON_USEDEP}]
dev-python/pyyaml[${PYTHON_USEDEP}]
dev-python/typing-extensions[${PYTHON_USEDEP}]
')
cuda? ( ~dev-libs/cutlass-3.9.2[tools(+)] )
onednn? ( sci-ml/ideep )
qnnpack? ( dev-libs/clog )
"
PATCHES=(
"${FILESDIR}"/${PN}-2.5.1-unbundle_fmt.patch
"${FILESDIR}"/${PN}-2.5.1-unbundle_kineto.patch
"${FILESDIR}"/${P}-unbundle_pocketfft.patch
"${FILESDIR}"/${PN}-2.5.1-cudnn_include_fix.patch
"${FILESDIR}"/${P}-gentoo.patch
"${FILESDIR}"/${PN}-2.4.0-cpp-httplib.patch
"${FILESDIR}"/${PN}-2.5.1-glog-0.6.0.patch
"${FILESDIR}"/${PN}-2.5.1-newfix-functorch-install.patch
"${FILESDIR}"/${PN}-2.6.0-rocm-fix-std-cpp17.patch
"${FILESDIR}"/${P}-cmake.patch
"${FILESDIR}"/${PN}-2.7.0-glog-0.7.1.patch
"${FILESDIR}"/${PN}-2.7.1-aotriton-fixes.patch
"${FILESDIR}"/${PN}-2.8.0-rocm-minus-flash.patch
)
src_prepare() {
if use flash || use memefficient; then
mv "${WORKDIR}"/${FLASH_P}/* third_party/${FLASH_PN}/ || die
fi
filter-lto #bug 862672
# Unbundle fmt
sed -i \
-e 's|::fmt-header-only||' \
c10/CMakeLists.txt \
cmake/Dependencies.cmake \
torch/CMakeLists.txt \
|| die
# Drop third_party from CMake tree
sed -i \
-e '/add_subdirectory.*third_party/d' \
CMakeLists.txt \
cmake/Dependencies.cmake \
cmake/ProtoBuf.cmake \
aten/src/ATen/CMakeLists.txt \
|| die
# Change libc10* path
sed -i \
-e "/EXPORT/s|DESTINATION lib)|DESTINATION $(get_libdir))|" \
c10/cuda/CMakeLists.txt \
c10/CMakeLists.txt \
c10/hip/CMakeLists.txt \
|| die
# Change libaotriton path
sed -i \
-e "s|}/lib|}/$(get_libdir)|g" \
cmake/External/aotriton.cmake \
|| die
# Noisy warnings from Logging.h
sed -i 's/-Wextra-semi//' cmake/public/utils.cmake || die
cmake_src_prepare
pushd torch/csrc/jit/serialization || die
flatc --cpp --gen-mutable --scoped-enums mobile_bytecode.fbs || die
popd
# prefixify the hardcoded paths, after all patches are applied
hprefixify \
aten/CMakeLists.txt \
caffe2/CMakeLists.txt \
cmake/Metal.cmake \
cmake/Modules/*.cmake \
cmake/Modules_CUDA_fix/FindCUDNN.cmake \
cmake/Modules_CUDA_fix/upstream/FindCUDA/make2cmake.cmake \
cmake/Modules_CUDA_fix/upstream/FindPackageHandleStandardArgs.cmake \
cmake/public/LoadHIP.cmake \
cmake/public/cuda.cmake \
cmake/Dependencies.cmake \
torch/CMakeLists.txt \
CMakeLists.txt
if use rocm; then
sed -e "s:/opt/rocm:/usr:" \
-e "s:lib/cmake:$(get_libdir)/cmake:g" \
-i cmake/public/LoadHIP.cmake || die
# TODO: delete, when caffe2 depends on systemwide composable_kernel
sed -e "s:third_party/composable_kernel:../composable_kernel-${CK_COMMIT}:g" \
-i aten/src/ATen/CMakeLists.txt || die
# Bug 959808: fix for gfx101x targets
pushd "${WORKDIR}/composable_kernel-${CK_COMMIT}" > /dev/null || die
eapply "${FILESDIR}"/composable-kernel-6.4.1-expand-isa.patch
popd > /dev/null || die
if tc-is-clang; then
# Systemwide gcc (for absl and at::TensorBase) + hipcc (llvm>=18) need abi-compat=17.
# But systemwide clang>=18 + hipcc (>=llvm-18) need opposite!
# See also: https://github.com/llvm/llvm-project/issues/102443#issuecomment-2329726287
sed '/-fclang-abi-compat=17/d' -i cmake/Dependencies.cmake || die
fi
# Workaround for libc++ issue https://github.com/llvm/llvm-project/issues/100802
sed 's/std::memcpy/memcpy/g' -i c10/util/Half.h || die
ebegin "HIPifying cuda sources"
${EPYTHON} tools/amd_build/build_amd.py || die
eend $?
fi
}
src_configure() {
if use cuda && [[ -z ${TORCH_CUDA_ARCH_LIST} ]]; then
ewarn "WARNING: caffe2 is being built with its default CUDA compute capabilities: 3.5 and 7.0."
ewarn "These may not be optimal for your GPU."
ewarn ""
ewarn "To configure caffe2 with the CUDA compute capability that is optimal for your GPU,"
ewarn "set TORCH_CUDA_ARCH_LIST in your make.conf, and re-emerge caffe2."
ewarn "For example, to use CUDA capability 7.5 & 3.5, add: TORCH_CUDA_ARCH_LIST=7.5 3.5"
ewarn "For a Maxwell model GPU, an example value would be: TORCH_CUDA_ARCH_LIST=Maxwell"
ewarn ""
ewarn "You can look up your GPU's CUDA compute capability at https://developer.nvidia.com/cuda-gpus"
ewarn "or by running /opt/cuda/extras/demo_suite/deviceQuery | grep 'CUDA Capability'"
fi
local mycmakeargs=(
-DBUILD_CUSTOM_PROTOBUF=OFF
-DLIBSHM_INSTALL_LIB_SUBDIR="${EPREFIX}"/usr/$(get_libdir)
-DPython_EXECUTABLE="${PYTHON}"
-DTORCH_INSTALL_LIB_DIR="${EPREFIX}"/usr/$(get_libdir)
-DUSE_CCACHE=OFF
-DUSE_CUDA=$(usex cuda)
-DUSE_DISTRIBUTED=$(usex distributed)
-DUSE_FAKELOWP=OFF
-DUSE_FBGEMM=$(usex fbgemm)
-DUSE_FLASH_ATTENTION=$(usex flash)
-DUSE_GFLAGS=ON
-DUSE_GLOG=ON
-DUSE_GLOO=$(usex gloo)
-DUSE_ITT=OFF
-DUSE_KINETO=ON
-DUSE_KLEIDIAI=OFF # TODO
-DUSE_MAGMA=OFF # TODO: In GURU as sci-libs/magma
-DUSE_MEM_EFF_ATTENTION=$(usex memefficient)
-DUSE_MKLDNN=$(usex onednn)
-DUSE_MPI=$(usex mpi)
-DUSE_NCCL=OFF
-DUSE_NNPACK=$(usex nnpack)
-DUSE_NUMA=OFF
-DUSE_NUMPY=$(usex numpy)
-DUSE_OPENCL=$(usex opencl)
-DUSE_OPENMP=$(usex openmp)
-DUSE_PYTORCH_QNNPACK=$(usex qnnpack)
-DUSE_PYTORCH_METAL=OFF
-DUSE_ROCM=$(usex rocm)
-DUSE_SYSTEM_CPUINFO=ON
-DUSE_SYSTEM_EIGEN_INSTALL=ON
-DUSE_SYSTEM_FP16=ON
-DUSE_SYSTEM_FXDIV=ON
-DUSE_SYSTEM_GLOO=ON
-DUSE_SYSTEM_NVTX=ON
-DUSE_SYSTEM_ONNX=ON
-DUSE_SYSTEM_PSIMD=ON
-DUSE_SYSTEM_PTHREADPOOL=ON
-DUSE_SYSTEM_PYBIND11=ON
-DUSE_SYSTEM_SLEEF=ON
-DUSE_SYSTEM_XNNPACK=$(usex xnnpack)
-DUSE_TENSORPIPE=$(usex distributed)
-DUSE_UCC=OFF
-DUSE_VALGRIND=OFF
-DUSE_XNNPACK=$(usex xnnpack)
-DUSE_XPU=OFF
-Wno-dev
)
if use mkl; then
mycmakeargs+=(-DBLAS=MKL)
elif use openblas; then
mycmakeargs+=(-DBLAS=OpenBLAS)
else
mycmakeargs+=(-DBLAS=Generic -DBLAS_LIBRARIES=)
fi
if use cuda; then
addpredict "/dev/nvidiactl" # bug 867706
addpredict "/dev/char"
addpredict "/proc/self/task" # bug 926116
mycmakeargs+=(
-DUSE_CUDNN=ON
-DTORCH_CUDA_ARCH_LIST="${TORCH_CUDA_ARCH_LIST:-3.5 7.0}"
-DUSE_NCCL=OFF # TODO: NVIDIA Collective Communication Library
-DCMAKE_CUDA_FLAGS="$(cuda_gccdir -f | tr -d \")"
-DUSE_CUSPARSELT=$(usex cusparselt)
)
elif use rocm; then
export PYTORCH_ROCM_ARCH="$(get_amdgpu_flags)"
if use memefficient; then
export AOTRITON_INSTALLED_PREFIX="${ESYSROOT}/usr"
fi
mycmakeargs+=(
-DUSE_NCCL=ON
-DUSE_SYSTEM_NCCL=ON
-DCMAKE_REQUIRE_FIND_PACKAGE_HIP=ON
)
# ROCm libraries produce too much warnings
append-cxxflags -Wno-deprecated-declarations -Wno-unused-result -Wno-unused-value
fi
if use onednn; then
mycmakeargs+=(
-DMKLDNN_FOUND=ON
-DMKLDNN_LIBRARIES=dnnl
-DMKLDNN_INCLUDE_DIR="${ESYSROOT}/usr/include/oneapi/dnnl"
)
fi
cmake_src_configure
}
src_compile() {
PYTORCH_BUILD_VERSION=${PV} \
PYTORCH_BUILD_NUMBER=0 \
cmake_src_compile
}
python_install() {
python_domodule python/torch
mkdir "${D}"$(python_get_sitedir)/torch/bin || die
mkdir "${D}"$(python_get_sitedir)/torch/lib || die
mkdir "${D}"$(python_get_sitedir)/torch/include || die
ln -s ../../../../../include/torch \
"${D}$(python_get_sitedir)"/torch/include/torch || die # bug 923269
ln -s ../../../../../bin/torch_shm_manager \
"${D}"/$(python_get_sitedir)/torch/bin/torch_shm_manager || die
ln -s ../../../../../$(get_libdir)/libtorch_global_deps.so \
"${D}"/$(python_get_sitedir)/torch/lib/libtorch_global_deps.so || die
}
src_install() {
cmake_src_install
# Used by pytorch ebuild
insinto "/var/lib/${PN}"
doins "${BUILD_DIR}"/CMakeCache.txt
dostrip -x /var/lib/${PN}/functorch.so
rm -rf python
mkdir -p python/torch || die
cp torch/version.py python/torch/ || die
python_install
}
@@ -1,141 +0,0 @@
Fix for "undeclared identifier 'CK_BUFFER_RESOURCE_3RD_DWORD'" for AMDGPU_TARGETS="gfx1012".
Combines of 3 patches from https://github.com/ROCm/composable_kernel/issues/775#issuecomment-2726315348
Bug: https://bugs.gentoo.org/947583
Bug: https://bugs.gentoo.org/show_bug.cgi?id=959808
--- a/include/ck/ck.hpp
+++ b/include/ck/ck.hpp
@@ -82,7 +82,7 @@ CK_DECLARE_ENV_VAR_BOOL(CK_LOGGING)
#define CK_BUFFER_RESOURCE_3RD_DWORD -1
#elif defined(__gfx803__) || defined(__gfx900__) || defined(__gfx906__) || defined(__gfx9__)
#define CK_BUFFER_RESOURCE_3RD_DWORD 0x00020000
-#elif defined(__gfx103__)
+#elif defined(__gfx101__) || defined(__gfx103__)
#define CK_BUFFER_RESOURCE_3RD_DWORD 0x31014000
#elif defined(__gfx11__) || defined(__gfx12__)
#define CK_BUFFER_RESOURCE_3RD_DWORD 0x31004000
@@ -90,12 +90,12 @@ CK_DECLARE_ENV_VAR_BOOL(CK_LOGGING)
// FMA instruction
#ifndef __HIP_DEVICE_COMPILE__ // for host code, define nothing
-#elif defined(__gfx803__) || defined(__gfx900__) // for GPU code
-#define CK_USE_AMD_V_MAC_F32
-#elif defined(__gfx906__) || defined(__gfx9__) || defined(__gfx103__) // for GPU code
+#elif defined(__gfx906__) || defined(__gfx9__) || defined(__gfx103__) || defined(__gfx1011__) || defined(__gfx1012__) // for GPU code
#define CK_USE_AMD_V_FMAC_F32
#define CK_USE_AMD_V_DOT2_F32_F16
#define CK_USE_AMD_V_DOT4_I32_I8
+#elif defined(__gfx803__) || defined(__gfx900__) || defined(__gfx101__) // for GPU code
+#define CK_USE_AMD_V_MAC_F32
#elif defined(__gfx11__) || defined(__gfx12__)
#define CK_USE_AMD_V_FMAC_F32
#define CK_USE_AMD_V_DOT2_F32_F16
--- a/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_multiple_d_dl.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_multiple_d_dl.hpp
@@ -71,7 +71,7 @@ __global__ void
const Block2CTileMap block_2_ctile_map)
{
#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx906__) || defined(__gfx908__) || \
- defined(__gfx90a__) || defined(__gfx94__) || defined(__gfx103__) || defined(__gfx11__) || \
+ defined(__gfx90a__) || defined(__gfx94__) || defined(__gfx101__) || defined(__gfx103__) || defined(__gfx11__) || \
defined(__gfx12__))
const index_t num_blocks_per_batch =
--- a/include/ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_dl.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_dl.hpp
@@ -51,7 +51,7 @@ __global__ void
const Block2CTileMap block_2_ctile_map)
{
#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx906__) || defined(__gfx9__) || \
- defined(__gfx103__) || defined(__gfx11__) || defined(__gfx12__))
+ defined(__gfx101__) || defined(__gfx103__) || defined(__gfx11__) || defined(__gfx12__))
constexpr index_t shared_block_size =
GridwiseGemm::GetSharedMemoryNumberOfByte() / sizeof(ABDataType);
--- a/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_weight_dl.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_weight_dl.hpp
@@ -48,7 +48,7 @@ __global__ void
const Block2CTileMap block_2_ctile_map,
const ComputePtrOffsetOfBatch compute_ptr_offset_of_batch)
{
-#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx906__) || defined(__gfx103__) || \
+#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx906__) || defined(__gfx101__) || defined(__gfx103__) || \
defined(__gfx90a__) || defined(__gfx908__) || defined(__gfx94__) || defined(__gfx11__) || \
defined(__gfx12__))
const index_t num_blocks_per_batch =
--- a/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_dl_multiple_d_nhwc_kyxc_nhwk.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_dl_multiple_d_nhwc_kyxc_nhwk.hpp
@@ -90,7 +90,7 @@ __global__ void
const Block2CTileMap block_2_ctile_map,
const ComputePtrOffsetOfBatch compute_ptr_offset_of_batch)
{
-#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx906__) || defined(__gfx103__) || \
+#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx906__) || defined(__gfx101__) || defined(__gfx103__) || \
defined(__gfx90a__) || defined(__gfx908__) || defined(__gfx94__) || defined(__gfx11__) || \
defined(__gfx12__))
// offset base pointer for each work-group
--- a/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_dl_nhwc_kyxc_nhwk.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_dl_nhwc_kyxc_nhwk.hpp
@@ -106,7 +106,7 @@ __global__ void
const Block2CTileMap block_2_ctile_map,
const ComputePtrOffsetOfBatch compute_ptr_offset_of_batch)
{
-#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx906__) || defined(__gfx103__) || \
+#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx906__) || defined(__gfx101__) || defined(__gfx103__) || \
defined(__gfx11__) || defined(__gfx12__))
// offset base pointer for each work-group
const index_t num_blocks_per_batch =
--- a/include/ck/tensor_operation/gpu/device/impl/device_grouped_gemm_multiple_d_dl.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_grouped_gemm_multiple_d_dl.hpp
@@ -40,7 +40,7 @@ __global__ void
const CDEElementwiseOperation cde_element_op)
{
#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx906__) || defined(__gfx908__) || \
- defined(__gfx90a__) || defined(__gfx103__) || defined(__gfx11__) || defined(__gfx94__) || \
+ defined(__gfx90a__) || defined(__gfx101__) || defined(__gfx103__) || defined(__gfx11__) || defined(__gfx94__) || \
defined(__gfx12__))
__shared__ char p_shared[GridwiseGemm::GetSharedMemoryNumberOfByte()];
--- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_dpp.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_dpp.hpp
@@ -28,7 +28,7 @@ __global__ void
#endif
kernel_gemm_dpp(const typename GridwiseGemm::Argument karg)
{
-#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx103__) || defined(__gfx11__))
+#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx101__) || defined(__gfx103__) || defined(__gfx11__))
__shared__ char p_shared[GridwiseGemm::GetSharedMemoryNumberOfByte()];
const auto a_grid_desc_ak0_m_ak1 = amd_wave_read_first_lane(
--- a/include/ck/tensor_operation/gpu/grid/gridwise_tensor_rearrange.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_tensor_rearrange.hpp
@@ -36,7 +36,7 @@ __global__ void
const ComputePtrOffsetOfStridedBatch compute_ptr_offset_of_batch)
{
#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx906__) || defined(__gfx908__) || \
- defined(__gfx90a__) || defined(__gfx94__) || defined(__gfx103__) || defined(__gfx11__) || \
+ defined(__gfx90a__) || defined(__gfx94__) || defined(__gfx101__) || defined(__gfx103__) || defined(__gfx11__) || \
defined(__gfx12__))
GridwiseTensorRearrangeKernel::Run(in_grid_desc,
p_in_global,
--- a/include/ck_tile/core/config.hpp
+++ b/include/ck_tile/core/config.hpp
@@ -10,6 +10,9 @@
#if defined(__gfx940__) || defined(__gfx941__) || defined(__gfx942__) || defined(__gfx950__)
#define __gfx94__
#endif
+#if defined(__gfx1010__) || defined(__gfx1011__) || defined(__gfx1012__)
+#define __gfx101__
+#endif
#if defined(__gfx1030__) || defined(__gfx1031__) || defined(__gfx1032__) || \
defined(__gfx1034__) || defined(__gfx1035__) || defined(__gfx1036__) || \
defined(__gfx10_3_generic__)
@@ -199,7 +202,7 @@
#elif defined(__gfx803__) || defined(__gfx900__) || defined(__gfx906__) || \
defined(__gfx9__) // for GPU code
#define CK_TILE_BUFFER_RESOURCE_3RD_DWORD 0x00020000
-#elif defined(__gfx103__) // for GPU code
+#elif defined(__gfx101__) || defined(__gfx103__) // for GPU code
#define CK_TILE_BUFFER_RESOURCE_3RD_DWORD 0x31014000
#elif defined(__gfx11__) || defined(__gfx12__) // for GPU code
#define CK_TILE_BUFFER_RESOURCE_3RD_DWORD 0x31004000