1 |
This eclass provides utilities for ROCm libraries in |
2 |
https://github.com/ROCmSoftwarePlatform, e.g. rocBLAS, rocFFT. |
3 |
It contains a USE_EXPAND, amdgpu_targets_*, which handles the GPU |
4 |
architecture to compile, and keep targets coherent among dependencies. |
5 |
Packages that depend on ROCm libraries, like cupy, can also make use of |
6 |
this eclass, mainly specify GPU architecture and it's corresponding |
7 |
dependencies via USE_EXPAND. |
8 |
|
9 |
Closes: https://bugs.gentoo.org/810619 |
10 |
Bugs: https://bugs.gentoo.org/817440 |
11 |
Signed-off-by: Yiyang Wu <xgreenlandforwyy@×××××.com> |
12 |
--- |
13 |
eclass/rocm.eclass | 284 ++++++++++++++++++++++++++++++++++++ |
14 |
profiles/base/make.defaults | 2 +- |
15 |
2 files changed, 285 insertions(+), 1 deletion(-) |
16 |
create mode 100644 eclass/rocm.eclass |
17 |
|
18 |
diff --git a/eclass/rocm.eclass b/eclass/rocm.eclass |
19 |
new file mode 100644 |
20 |
index 000000000000..1866d6b7cc94 |
21 |
--- /dev/null |
22 |
+++ b/eclass/rocm.eclass |
23 |
@@ -0,0 +1,284 @@ |
24 |
+# Copyright 2022 Gentoo Authors |
25 |
+# Distributed under the terms of the GNU General Public License v2 |
26 |
+ |
27 |
+# @ECLASS: rocm.eclass |
28 |
+# @MAINTAINER: |
29 |
+# Gentoo Science Project <sci@g.o> |
30 |
+# @AUTHOR: |
31 |
+# Yiyang Wu <xgreenlandforwyy@×××××.com> |
32 |
+# @SUPPORTED_EAPIS: 7 8 |
33 |
+# @BLURB: Common functions and variables for ROCm packages written in HIP |
34 |
+# @DESCRIPTION: |
35 |
+# ROCm packages such as sci-libs/<roc|hip>* can utilize functions in this |
36 |
+# eclass. Currently, it handles the AMDGPU_TARGETS variable via USE_EXPAND, so |
37 |
+# user can use USE flag to control which GPU architecture to compile, and |
38 |
+# ensure coherence among dependencies. It also specify CXX=hipcc, to let hipcc |
39 |
+# compile. Another important function is src_test, which checks whether a valid |
40 |
+# KFD device exists for testing, and then execute the test program. |
41 |
+# |
42 |
+# Most ROCm packages use cmake as build system, so this eclass does not export |
43 |
+# phase functions which overwrites the phase functions in cmake.eclass. Ebuild |
44 |
+# should explicitly call rocm-{configure,test} in src_configure and src_test. |
45 |
+# |
46 |
+# @EXAMPLE: |
47 |
+# @CODE |
48 |
+# # Example ebuild for ROCm library in https://github.com/ROCmSoftwarePlatform |
49 |
+# # whcih depends on rocBLAS |
50 |
+# inherit cmake rocm |
51 |
+# # ROCm libraries SRC_URI is usually in form of: |
52 |
+# SRC_URI="https://github.com/ROCmSoftwarePlatform/${PN}/archive/rocm-${PV}.tar.gz -> ${P}.tar.gz" |
53 |
+# S=${WORKDIR}/${PN}-rocm-${PV} |
54 |
+# SLOT="0/$(ver_cut 1-2)" |
55 |
+# IUSE="test" |
56 |
+# REQUIRED_USE="${ROCM_REQUIRED_USE}" |
57 |
+# RESTRICT="!test? ( test )" |
58 |
+# |
59 |
+# RDEPEND=" |
60 |
+# dev-util/hip |
61 |
+# sci-libs/rocBLAS:${SLOT}[${ROCM_USEDEP}] |
62 |
+# " |
63 |
+# |
64 |
+# src_configure() { |
65 |
+# local mycmakeargs=( |
66 |
+# -DBUILD_CLIENTS_TESTS=$(usex test ON OFF) |
67 |
+# ) |
68 |
+# rocm-configure |
69 |
+# } |
70 |
+# |
71 |
+# src_test() { |
72 |
+# rocm-test |
73 |
+# } |
74 |
+# @CODE |
75 |
+# |
76 |
+# # Example for packages depend on ROCm libraries -- a package depend on |
77 |
+# # rocBLAS, and use comma seperated ${HCC_AMDGPU_TARGET} to determine GPU |
78 |
+# # architecture to compile. Requires ROCm version >5. |
79 |
+# @CODE |
80 |
+# ROCM_VERSION=5.1 |
81 |
+# inherit rocm |
82 |
+# IUSE="rocm" |
83 |
+# REQUIRED_USE="rocm? ( ${ROCM_REQUIRED_USE} )" |
84 |
+# DEPEND="rocm? ( >=dev-util/hip-${ROCM_VERSION} |
85 |
+# >=sci-libs/rocBLAS-${ROCM_VERSION}[${ROCM_USEDEP}] )" |
86 |
+# .... |
87 |
+# src_configure() { |
88 |
+# if use rocm; then |
89 |
+# local AMDGPU_FLAGS=$(get_amdgpu_flags) |
90 |
+# export HCC_AMDGPU_TARGET=${AMDGPU_FLAGS//;/,} |
91 |
+# fi |
92 |
+# default |
93 |
+# } |
94 |
+# @CODE |
95 |
+ |
96 |
+if [[ ! ${_ROCM_ECLASS} ]]; then |
97 |
+ |
98 |
+case ${EAPI} in |
99 |
+ 7|8) ;; |
100 |
+ *) die "${ECLASS}: EAPI ${EAPI:-0} not supported" ;; |
101 |
+esac |
102 |
+ |
103 |
+inherit edo |
104 |
+ |
105 |
+# @ECLASS_VARIABLE: ROCM_VERSION |
106 |
+# @DEFAULT_UNSET |
107 |
+# @PRE_INHERIT |
108 |
+# @DESCRIPTION: |
109 |
+# The ROCm version of current package. Default is ${PV}, but for other packages |
110 |
+# that depend on ROCm libraries, this can be set to match the version of |
111 |
+# required ROCm libraries. |
112 |
+ |
113 |
+# @ECLASS_VARIABLE: ALL_AMDGPU_TARGETS |
114 |
+# @INTERNAL |
115 |
+# @DESCRIPTION: |
116 |
+# The list of USE flags corresponding to all AMDGPU targets in this ROCm |
117 |
+# version. The value depends on ${PV}. Architectures and devices map: |
118 |
+# https://www.coelacanth-dream.com/posts/2019/12/30/did-rid-product-matome-p2 |
119 |
+ |
120 |
+# @ECLASS_VARIABLE: OFFICIAL_AMDGPU_TARGETS |
121 |
+# @INTERNAL |
122 |
+# @DESCRIPTION: |
123 |
+# The list of USE flags corresponding to all officially supported AMDGPU |
124 |
+# targets in this ROCm version, documented at |
125 |
+# https://docs.amd.com/bundle/ROCm-Installation-Guide-v${PV}/page/Prerequisite_Actions.html. |
126 |
+# USE flag of these architectures will be default on. Depends on ${PV}. |
127 |
+ |
128 |
+# @ECLASS_VARIABLE: ROCM_REQUIRED_USE |
129 |
+# @OUTPUT_VARIABLE |
130 |
+# @DESCRIPTION: |
131 |
+# Requires at least one AMDGPU target to be compiled. |
132 |
+# Example use for ROCm libraries: |
133 |
+# @CODE |
134 |
+# REQUIRED_USE="${ROCM_REQUIRED_USE}" |
135 |
+# @CODE |
136 |
+# Example use for packages that depend on ROCm libraries |
137 |
+# @CODE |
138 |
+# IUSE="rocm" |
139 |
+# REQUIRED_USE="rocm? ( ${ROCM_REQUIRED_USE} )" |
140 |
+# @CODE |
141 |
+ |
142 |
+# @ECLASS_VARIABLE: ROCM_USEDEP |
143 |
+# @OUTPUT_VARIABLE |
144 |
+# @DESCRIPTION: |
145 |
+# This is an eclass-generated USE-dependency string which can be used to |
146 |
+# depend on another ROCm package being built for the same AMDGPU architecture. |
147 |
+# |
148 |
+# The generated USE-flag list is compatible with packages using rocm.eclass. |
149 |
+# |
150 |
+# Example use: |
151 |
+# @CODE |
152 |
+# DEPEND="sci-libs/rocBLAS[${ROCM_USEDEP}]" |
153 |
+# @CODE |
154 |
+ |
155 |
+# @FUNCTION: _rocm_set_globals |
156 |
+# @DESCRIPTION: |
157 |
+# Set global variables used by the eclass: ALL_AMDGPU_TARGETS, |
158 |
+# OFFICIAL_AMDGPU_TARGETS, ROCM_REQUIRED_USE, and ROCM_USEDEP |
159 |
+_rocm_set_globals() { |
160 |
+ case ${ROCM_VERSION:-${PV}} in |
161 |
+ 4.*) |
162 |
+ ALL_AMDGPU_TARGETS=( |
163 |
+ gfx803 gfx900 gfx906 gfx908 |
164 |
+ gfx1010 gfx1011 gfx1012 gfx1030 |
165 |
+ ) |
166 |
+ OFFICIAL_AMDGPU_TARGETS=( |
167 |
+ gfx906 gfx908 |
168 |
+ ) |
169 |
+ ;; |
170 |
+ 5.*) |
171 |
+ ALL_AMDGPU_TARGETS=( |
172 |
+ gfx803 gfx900 gfx906 gfx908 gfx90a |
173 |
+ gfx1010 gfx1011 gfx1012 gfx1030 gfx1031 |
174 |
+ ) |
175 |
+ OFFICIAL_AMDGPU_TARGETS=( |
176 |
+ gfx906 gfx908 gfx90a gfx1030 |
177 |
+ ) |
178 |
+ ;; |
179 |
+ *) |
180 |
+ die "Unknown ROCm major version! Please update rocm.eclass before bumping to new ebuilds" |
181 |
+ ;; |
182 |
+ esac |
183 |
+ |
184 |
+ ROCM_REQUIRED_USE+=" || (" |
185 |
+ for gpu_target in "${ALL_AMDGPU_TARGETS[@]}"; do |
186 |
+ if has "${gpu_target}" "${OFFICIAL_AMDGPU_TARGETS[@]}"; then |
187 |
+ IUSE+=" ${gpu_target/#/+amdgpu_targets_}" |
188 |
+ else |
189 |
+ IUSE+=" ${gpu_target/#/amdgpu_targets_}" |
190 |
+ fi |
191 |
+ ROCM_REQUIRED_USE+=" ${gpu_target/#/amdgpu_targets_}" |
192 |
+ done |
193 |
+ ROCM_REQUIRED_USE+=" ) " |
194 |
+ |
195 |
+ local flags=( "${ALL_AMDGPU_TARGETS[@]/#/amdgpu_targets_}" ) |
196 |
+ local optflags=${flags[@]/%/(-)?} |
197 |
+ ROCM_USEDEP=${optflags// /,} |
198 |
+} |
199 |
+_rocm_set_globals |
200 |
+unset -f _rocm_set_globals |
201 |
+ |
202 |
+ |
203 |
+# @FUNCTION: get_amdgpu_flags |
204 |
+# @USAGE: get_amdgpu_flags |
205 |
+# @DESCRIPTION: |
206 |
+# Convert specified use flag of amdgpu_targets to compilation flags. |
207 |
+# Append default target feature to GPU arch. See |
208 |
+# https://llvm.org/docs/AMDGPUUsage.html#target-features |
209 |
+get_amdgpu_flags() { |
210 |
+ local AMDGPU_TARGET_FLAGS |
211 |
+ for gpu_target in "${ALL_AMDGPU_TARGETS[@]}"; do |
212 |
+ local target_feature= |
213 |
+ if use amdgpu_targets_${gpu_target}; then |
214 |
+ case ${gpu_target} in |
215 |
+ gfx906|gfx908) |
216 |
+ target_feature=:xnack- |
217 |
+ ;; |
218 |
+ gfx90a) |
219 |
+ target_feature=:xnack+ |
220 |
+ ;; |
221 |
+ *) |
222 |
+ ;; |
223 |
+ esac |
224 |
+ AMDGPU_TARGET_FLAGS+="${gpu_target}${target_feature};" |
225 |
+ fi |
226 |
+ done |
227 |
+ echo ${AMDGPU_TARGET_FLAGS} |
228 |
+} |
229 |
+ |
230 |
+# @FUNCTION: check_rw_permission |
231 |
+# @USAGE: check_rw_permission <file> |
232 |
+# @DESCRIPTION: |
233 |
+# check read and write permissions on a specific file, die if no permission. |
234 |
+# @EXAMPLE: |
235 |
+# @CODE |
236 |
+# check_rw_permission /dev/kfd |
237 |
+# CODE |
238 |
+check_rw_permission() { |
239 |
+ if [[ ! -r $1 ]] || [[ ! -w $1 ]]; then |
240 |
+ eerror "Portage do not have read or write permissions on $1!" |
241 |
+ eerror "Make sure both are in render group and check the permissions." |
242 |
+ die "No permissions on $1" |
243 |
+ fi |
244 |
+} |
245 |
+ |
246 |
+# == phase functions == |
247 |
+ |
248 |
+# @FUNCTION: rocm-configure |
249 |
+# @DESCRIPTION: |
250 |
+# configure rocm packages, and setting common cmake arguments. Only for ROCm |
251 |
+# libraries in https://github.com/ROCmSoftwarePlatform using cmake. |
252 |
+rocm-configure() { |
253 |
+ # avoid sandbox violation |
254 |
+ addpredict /dev/kfd |
255 |
+ addpredict /dev/dri/ |
256 |
+ |
257 |
+ mycmakeargs+=( |
258 |
+ -DAMDGPU_TARGETS="$(get_amdgpu_flags)" |
259 |
+ -DCMAKE_SKIP_RPATH=TRUE |
260 |
+ ) |
261 |
+ |
262 |
+ CXX="hipcc" cmake_src_configure |
263 |
+} |
264 |
+ |
265 |
+# @FUNCTION: rocm-test |
266 |
+# @DESCRIPTION: |
267 |
+# Test whether valid GPU device is present. If so, execute test. |
268 |
+# @EXAMPLE: |
269 |
+# ROCm packages can have two test scenarioes: |
270 |
+# 1. cmake_src_test. MAKEOPTS="-j1" ensures only one test on GPU at a time; |
271 |
+# @CODE |
272 |
+# LD_LIBRARY_PATH=<path-to-lib> rocm-test --cmake |
273 |
+# @CODE |
274 |
+# 2. one gtest binary called "${PN,,}"-test in ${BUILD_DIR}/clients/staging; |
275 |
+# @CODE |
276 |
+# cd "${BUILD_DIR}"/clients/staging || die |
277 |
+# LD_LIBRARY_PATH=<path-to-lib> rocm-test "${PN,,}"-test |
278 |
+# @CODE |
279 |
+# Some packages like rocFFT have two test binaries like rocfft-selftest; |
280 |
+# packages like dev-libs/rccl have test binary with custom names. |
281 |
+# @CODE |
282 |
+# cd "${BUILD_DIR}"/clients/staging || die |
283 |
+# export LD_LIBRARY_PATH=<path-to-lib> |
284 |
+# cd <test-bin-location> || die |
285 |
+# rocm-test <test-bin-1> |
286 |
+# rocm-test <test-bin-2> |
287 |
+# @CODE |
288 |
+rocm-test() { |
289 |
+ # grant and check permissions on /dev/kfd and /dev/dri/render* |
290 |
+ for device in /dev/kfd /dev/dri/render*; do |
291 |
+ addwrite "${device}" |
292 |
+ check_rw_permission "${device}" |
293 |
+ done |
294 |
+ |
295 |
+ case ${1} in |
296 |
+ --cmake) |
297 |
+ # Avoid multi jobs running that may cause GPU error or CPU overload |
298 |
+ MAKEOPTS="-j1" cmake_src_test |
299 |
+ ;; |
300 |
+ *) |
301 |
+ edob ./${1} |
302 |
+ ;; |
303 |
+ esac |
304 |
+} |
305 |
+ |
306 |
+_ROCM_ECLASS=1 |
307 |
+fi |
308 |
diff --git a/profiles/base/make.defaults b/profiles/base/make.defaults |
309 |
index 326cb28de537..2c288d12d103 100644 |
310 |
--- a/profiles/base/make.defaults |
311 |
+++ b/profiles/base/make.defaults |
312 |
@@ -13,7 +13,7 @@ USE_EXPAND_VALUES_USERLAND="BSD GNU" |
313 |
|
314 |
# Env vars to expand into USE vars. Modifying this requires prior |
315 |
# discussion on gentoo-dev@l.g.o. |
316 |
-USE_EXPAND="ABI_MIPS ABI_S390 ABI_X86 ADA_TARGET ALSA_CARDS APACHE2_MODULES APACHE2_MPMS CALLIGRA_FEATURES CAMERAS COLLECTD_PLUGINS CPU_FLAGS_ARM CPU_FLAGS_PPC CPU_FLAGS_X86 CURL_SSL ELIBC FFTOOLS GPSD_PROTOCOLS GRUB_PLATFORMS INPUT_DEVICES KERNEL L10N LCD_DEVICES LIBREOFFICE_EXTENSIONS LLVM_TARGETS LUA_SINGLE_TARGET LUA_TARGETS MONKEYD_PLUGINS NGINX_MODULES_HTTP NGINX_MODULES_MAIL NGINX_MODULES_STREAM OFFICE_IMPLEMENTATION OPENMPI_FABRICS OPENMPI_OFED_FEATURES OPENMPI_RM PHP_TARGETS POSTGRES_TARGETS PYTHON_SINGLE_TARGET PYTHON_TARGETS QEMU_SOFTMMU_TARGETS QEMU_USER_TARGETS ROS_MESSAGES RUBY_TARGETS SANE_BACKENDS USERLAND UWSGI_PLUGINS VIDEO_CARDS VOICEMAIL_STORAGE XTABLES_ADDONS" |
317 |
+USE_EXPAND="ABI_MIPS ABI_S390 ABI_X86 ADA_TARGET ALSA_CARDS AMDGPU_TARGETS APACHE2_MODULES APACHE2_MPMS CALLIGRA_FEATURES CAMERAS COLLECTD_PLUGINS CPU_FLAGS_ARM CPU_FLAGS_PPC CPU_FLAGS_X86 CURL_SSL ELIBC FFTOOLS GPSD_PROTOCOLS GRUB_PLATFORMS INPUT_DEVICES KERNEL L10N LCD_DEVICES LIBREOFFICE_EXTENSIONS LLVM_TARGETS LUA_SINGLE_TARGET LUA_TARGETS MONKEYD_PLUGINS NGINX_MODULES_HTTP NGINX_MODULES_MAIL NGINX_MODULES_STREAM OFFICE_IMPLEMENTATION OPENMPI_FABRICS OPENMPI_OFED_FEATURES OPENMPI_RM PHP_TARGETS POSTGRES_TARGETS PYTHON_SINGLE_TARGET PYTHON_TARGETS QEMU_SOFTMMU_TARGETS QEMU_USER_TARGETS ROS_MESSAGES RUBY_TARGETS SANE_BACKENDS USERLAND UWSGI_PLUGINS VIDEO_CARDS VOICEMAIL_STORAGE XTABLES_ADDONS" |
318 |
|
319 |
# USE_EXPAND variables whose contents are not shown in package manager |
320 |
# output. Changes need discussion on gentoo-dev. |
321 |
-- |
322 |
2.34.1 |